Centralized Institutions and Sudden Change�

JARED RUBINy

California State University, Fullerton

First Draft: July 2009

This Draft: March 2010

Abstract

Why do sudden and massive social, economic, and political changes occur when and where they

do? Are there institutional preconditions that encourage such changes when present and discourage such

changes when absent? In this paper, I employ a general model which suggests that such changes are more

likely to occur in regimes with centralized coercive power - those with the ability to impose more than one

type of sanction (economic, legal, political, social, or religious). On the one hand, centralized authorities

are often better able to suppress subversive actions by the citizenry when external shocks are small or do

not affect many, as citizens have little incentive to incur numerous types of sanctions. However, citizens in

such regimes are also more likely to lie about their internal preferences (e.g., falsely declare loyalty to an

oppressive government), and authorities have less incentive to accommodate dissent than in decentralized

regimes. Yet, this entails that large shocks which encourage some individuals to transgress the authorities'

dictates may trigger a cascade - as more individuals change actions, social pressures encourage even more

to change, and a new, substantively different outcome emerges. The model is applied to the occurrence and

severity of protests that followed austerity measures taken in developing nations since the 1970s as well as

the Iranian uprising of summer 2009.

�I am extremely grateful to Avner Greif, Kristin Kleinjans, and Timur Kuran for their extremely helpful insights. Comments received

at seminars at Cal State Fullerton were also quite helpful. All errors are mine.

yE-mail: jrubin@fullerton.edu; Address: Department of Economics, California State University, 800 N. State College Blvd., Fuller-

ton, CA 92834; Tel: (657) 278-2387; Fax: (657) 278-3097; http://faculty.fullerton.edu/jrubin/

1 Introduction

Economists are well aware that small events may act as a spark that leads to a signi�cant change in equi-

librium outcomes. In some instances such events act as a signal in the decision process, leading to social

learning phenomena such as herd behavior and informational cascades that encourage individuals to ignore

their private information and follow the actions of others (Banerjee 1992, 1993; Bikhchandani, Hirshleifer,

and Welch 1992, 1998).1 In other cases, small events encourage some individuals to publicly reveal their pre-

viously suppressed, privately held preferences, which leads to information revelation and changes in status

or reputation bestowed on certain actions, in turn resulting in starkly different equilibria (Kuran 1989, 1995a,

1995b, 1998; Bernheim 1994; Kuran and Sunstein 1999; Kuran and Sandholm 2008).

These works contribute signi�cantly to our understanding of social behavior and help explain why mas-

sive social change can occur. However, this literature lacks an account of the broad institutional and macro-

economic settings under which massive social change arises - why do we see such change under certain

circumstances but not others?2

This paper provides such an account, suggesting that massive economic, social, or political change may

be one consequence of an institutional arrangement in which one type of authority (economic, legal, political,

social, or religious) has multi-lateral coercive power. I formulate a model with three types of agents: a group

of citizens and two different institutional authorities. The latter players can be thought of as religious authori-

ties, political �gures, social icons, or legal authorities. One of the authorities is a "centralized" authority, with

the degree of centralization being the level of the cost imposed on the other authority for transgressing the

central authority's dictates. This speci�cation re�ects the idea that political authorities often depend on reli-

gious authorities for legitimacy, economic authorities are often subject to the whims of political authorities,

and the like.

1These phenomena have also been studied in the experimental laboratory. See Anderson and Holt (1997), Çelen and Kariv (2004).

2For example, Banerjee (1992, 1993) provides a basic set of circumstances that lead to herd behavior, but provides no macroeconomic

or institutional conditions under which such behavior is more likely to arise. Similarly, Bikhchandani, Hirshleifer, and Welch (1992,

1998) do not incorporate such conditions into their study of information cascades. Kuran (1989) emphasizes that a pre-condition for

revolutionary activity can be a substantial amount of private opposition (and public support) for a regime, but does not specify why such

conditions arise. Kuran (1995a) does note that preference falsi�cation is less likely to happen in democracy than in autocracy because

the former has fewer means of restricting expression, but he does not provide an explanation for why this is so. Bernheim (1994) hints

at a cultural explanation, arguing that this phenomenon is more probable in societies that place more emphasis on social status.

1

The model analyzes the interactions between citizens and the authorities under the situation where the

preferences of the latter are not aligned with those of the former. Citizens derive intrinsic utility from their

own actions and have sanctions imposed on them for diverging from the actions of the two institutional

authorities. These "actions" could represent any number of phenomena in which the desires of some citizens

diverge from those of institutional authorities, such as speaking or writing freely, having more than one child,

or practicing minority religions. Citizens also derive disutility when their actions differ from endogenously

determined social norms, which may arise from the importance individuals place on social identity (Akerlof

and Kranton 2000, 2005), social custom and reputation (Akerlof 1980; Romer 1984; Kuran 1989, 1998;

Kuran and Sunstein 1999), or status and conformity (Fershtman and Weiss 1993; Bernheim 1994; Fershtman,

Murphy, and Weiss 1996; Akerlof 1997; Kuran and Sandholm 2008).

Unlike works on preference falsi�cation (especially Kuran [1987, 1989, 1995a, 1998]), I do not assume

that citizens' optimal actions are constant over time in the absence of social sanctions or preference updating.

Instead, the institutional authorities respond to the actions of the citizenry, which in turn changes citizens'

optimal actions. These dynamics entail the result that highly-centralized authorities - those with coercive

power over more than one type of sanction - are more insulated from pressures for change when exogenous

shocks are small. In this case, there is less incentive for citizens in centralized economies to violate the

dictates of the authorities, as they incur more than one type of sanction from doing so. In turn, actions remain

stable as long as shocks are suf�ciently small.

However, this logic also entails that citizens are more likely to choose actions further away from their

intrinsic optimum (or bliss point) in centralized economies. Moreover, centralized authorities have less incen-

tive to accommodate the actions of the citizenry than decentralized authorities, since decentralized authorities

can respond to their actions without fear of censure from a central authority. These two features entail that

a cascade or bandwagon effect may result in centralized regimes when a shock arises that is large enough

to encourage some citizens to transgress the authorities' dictates. In other words, when a small portion of

society transgresses the law, the publicly revealed actions of these citizens changes the social norm so that

more citizens (who choose actions diverging from their intrinsic optimum) are encouraged to transgress the

law, which encourages even more to evade the law, and so on. This social effect, in turn, encourages the

authorities to signi�cantly update their dictates, resulting in massive change from the previous equilibrium.

This is not the case in less centralized economies, as authorities are more likely to accommodate the actions

of the citizenry, and preferences are therefore less likely to be falsi�ed.

2

One implication of this thesis is that institutional authorities with multi-lateral coercive power may seem

insulated from upheaval when in fact they are quite vulnerable. Because of this underlying vulnerability,

authorities in such societies often do anything to suppress large shocks from occurring, such as suppressing

and controlling media and harshly combating dissent. Such tactics work to suppress small shocks. However,

this also works to further push the actions of the citizenry away from their preferences, which may cause a

larger cascade and thus unintentionally sow the seeds of the authority's demise.3

This paper relates to the large economic, sociological, and political science literature in revolutions.4

The present paper is not meant to accept or deny the validity of any of these arguments, but instead offers

a complementary hypothesis. I do not account for the organizations or leadership that is often instrumental

to revolutionary activity, but instead provide a mapping from broad institutional structure to massive social,

political, and economic change.

Historical examples of massive and unexpected changes occurring in centralized regimes include the

Iranian Revolution of 1979, the Bolshevik Revolution of 1917, the Taiping Rebellion in China (1850-1864),

and the Protestant Reformation. In this paper, I employ the insights of the model to analyze the numerous

austerity protests which occurred in the developing world since the 1970s as well as the Iranian uprising of

summer 2009. An econometric analysis of these protests from 1976-1992 suggests that protests were more

severe in decentralized economies if they followed small shocks (proxied for by indices of IMF involve-

ment), but were more severe in centralized economies if they followed large shocks. Moreover, an analytical

narrative of the events following the disputed Iranian election of June 2009 suggests that the severity of the

protests following the "shock" of the election results was a direct consequence of the institutional situation in

which political authorities (especially President Ahmadinejad) depended on religious authorities (especially

Supreme Leader Ayatollah Ali Khamenei) for legitimacy, giving the latter a great deal of control over all

types of sanctions - religious, political, social, legal, and economic.

3For more on the broad effects of the centralization of coercive power throughout world history, see Greif (2005). Iyigun (2009) argues

that there is a positive connection between monotheism and the length and bredth of dynastic power, as monotheisitc religions have

general been complementary to centralized government (due to high �xed costs of starting a monotheistic religion). This argument is

quite consistent with the one made in the present paper.

4The mechanism underlying revolutionary activity in the present paper is closest to Kuran (1989, 1995a, 1995b), who analyzes the

implications of public revelations of private preferences. For overviews of the social analysis of revolutions, see Tanter and Midlarsky

(1967), Shugart (1989), and Goldstone (1994, 2001).

3

2 The Model

2.1 Setup

Consider an economy in which all players are in�nitely lived and actions are common knowledge. The

players engage in a dynamic game with perfect information of previous actions. There are N + 2 players

(where N is large): N heterogeneous citizens and two institutional (social, political, economic, legal, or

religious) authorities, I1 and I2.5

Citizens are conformists and derive utility in three ways: from choosing actions close to their intrinsic

"bliss point", conforming to endogenously determined social norms, and conforming to the actions of the

institutional authorities.6 The model analyzes situations in which the preferences of some citizens differ

exogenously from those of the authorities, so actions could represent varying levels of freedom of speech,

press, or religion, publicly expressed dissatisfaction with the government or religious authorities, or public

opinion on social issues.

Each period (denoted with subscript t) begins with every citizen j choosing some observable action,

aj;t 2 R+j . They have unobserved types, aBj 2 R+j , which represent their exogenously determined intrinsic

bliss points. Their intrinsic utility is decreasing in the distance between their action and their type.7 After

citizens act, authorities I1 and I2 choose observable actions i1;t 2 R+ and i2;t 2 R+. The endogenization

of this choice differs from previous papers focusing on institutional focal points, such as Kuran (1987, 1989,

1995a, 1998), which assume that such points are exogenously given.8 I1 and I2 have unobserved types,

iB1 2 R+ and iB2 2 R+, and their utility is decreasing in the distance between their action and their type.

Citizens face three costs. Two of these costs are a function of the distance between the citizen's action

and the actions of the two authorities. These costs are increasing in the size of the violation (as in Romer

5The inclusion of only two authorities allows for tractability. The intuition underlying the main results of the model holds in more realistic

situations including numerous types of authorities.

6The speci�cation in the present model assumes that individuals derive utility from conforming. Gints (2003) suggests that "pro-social"

behavior may be biologically determined, as humans improved their biological "�tness" by internalizing cultural norms. See also Greif

(2009, 2010). A richer model would include a portion of non-conformists and would arrive at similar results as long as this portion is

not too large or heavily concentrated in the central portion of the type distribution.

7This formulation is similar to those in Bernheim (1994) and Kuran (1987, 1995a, 1998). Like my model, these models show that many

individuals will conform to single actions despite heterogeneous underlying preferences.

8Greif (2009, 2010), however, proposes a model in which "moral authorities" choose norms in a manner similar to the one in this model.

4

[1984], Iannaccone [1988], Bernheim [1994], Kuran [1987, 1995a], Akerlof [1997], Kuran and Sandholm

[2008], Rubin [2009], and many others) and represent the costs associated with breaking a religious dictate,

breaking a law, violating a political norm, and the like, depending on the type of authority in question.

The third cost is a function of the distance between their action and the average action of the citizenry,

at =1N

Pj aj;t. This "social cost" captures the in�uence of social norms which may arise from the impor-

tance individuals place on social identity, social custom, reputation, status, or conformity. Assume that N is

large enough that each citizen has a trivial effect on at, so they view at as exogenously given. The salient

features of the norm are that it is endogenously determined and it is not affected by the actions of any one

citizen. The norm evolves over time whenever citizens change their actions.9

Each citizen j maximizes the following utility function in each period:

Uj;t = �b�aj;t � aBj

�2 � c (aj;t � i1;t�1)2 � d (aj;t � i2;t�1)2 � e (aj;t � at�1)2 , (1)

where b, c, d, and e are weighting parameters greater than zero.

The primary exogenous parameter of concern in the model, , denotes the degree to which authority I1

derives utility by conforming to the action of I2. Although there are certainly endogenous elements to

in reality - for example, the degree to which one authority depends on the other could be a function of the

degree to which the citizens abide by the dictates of the latter - endogenizing this key variable unnecessarily

complicates the model. This paper concentrates on massive equilibrium changes over a short period and

how such changes arise quite rapidly. Broader institutional change which endogenously effect the level of

centralization likely follows in the long run due to the interactions described in the model, but should not be

affected in the short run by the rapid, massive change which is at the heart of this model.10

measures the degree to which I2 has multi-lateral coercive power over the citizenry, as higher levels of

indicate that I2 has greater in�uence over the sanctions imposed by I1. Centralization of coercive power

may arise from political authorities depending on religious authorities for legitimacy, economic authorities

9The social cost accounts for numerous utility speci�cations, such as one in which citizens derive utility from minimizing the sum of

the difference between their actions and each of the citizenry. Alternatively, social costs could be modeled as resulting from divergence

from the mode action of the citizenry. I do not employ this speci�cation because it is less tractable but it would not alter the intuition

espoused in the model.

10Moreover, Rubin (2009) shows in a similar model that under a basic set of circumstances, endogenizing the degree of centralization

merely exacerbates the effects seen under an exogenous parameterization, as the feedback between players is more enhanced.

5

depending on political or legal authorities for long-term viability, and the like.11 The centralization parameter

enters I1's utility as a scalar that affects the utility I1 receives from choosing an action different from I2. At

= 0, there is no such cost, at = 1, there is an in�nite cost (if i1;t 6= i2;t�1), and at 2 (0;1), there

is a positive cost that is increasing in . While it is possible (in reality) that I2 may also face costs from not

conforming to I1, the assumption of unidirectional centralization of coercive power allows for tractability. It

follows from this setup that at =1, the two authorities are ostensibly the same actor: power over numerous

dimensions is centralized in one authority (I2). At large , I2 has signi�cant but not unlimited power over

varying types of sanctions.

I1 faces a public support cost which is increasing in the distance between its action and the average action

of the citizenry. This speci�cation accounts for numerous factors underlying the relationship between insti-

tutional authorities and the citizenry. For one, the authority's legitimacy is undermined when its dictates are

transgressed by society. Moreover, transgression by the citizenry signals that the dictate decreases ef�ciency,

which would in turn decrease the utility of the authority if it is in a position to extract rents from the citizenry

(Rubin 2009). The upshot of this speci�cation is that institutional laws are endogenously determined.

I1 also derives utility from minimizing the distance between its action and its own internal bliss point

(iB1 ), and it maximizes the following utility function in each period:

V 1t = �f�i1;t � iB1

�2 � g (i1;t � i2;t�1)2 � h (i1;t � at)2 ; (2)

where f , g, and h are weighting parameters greater than zero.

I2 derives utility in a similar manner as I1, except that it is not concerned with conforming to the action

of the other authority. It maximizes the following utility function in each period:

V 2t = �j�i2;t � iB2

�2 � k (i2;t � at)2 , (3)

where j, and k are weighting parameters greater than zero. Since the model is intended to shed light on the

consequences of centralization of institutional authority, I assume that I2 is more concerned with conforming

to its bliss point than the mean action of the citizenry relative to I1. In other words, assume that jh > kf . I

11For more on the role of the interaction between political and religious institutions in the Islamic and Christian worlds, see Lewis (1974,

1993, 1995), Razi (1987), Greif (2002), Kuran (2005), Rubin (2009), Cosgel, Miceli, and Ahmed (2009), Cosgel and Miceli (2009), and

Cosgel, Miceli, and Rubin (2010a, 2010b). For more on the centralization of coercive power, see Greif (2005) and Karaman (2009).

6

also assume that both institutional authorities derive more utility from actions close to their bliss points than

from actions close to the mean action of the citizenry. That is, j > k and f > h. Dropping these assumptions

does not dramatically change the results but does entail that they do not hold for a small (and unrealistic) part

of the parameter set.

2.2 Institutions and Massive Social Change

In this section I study behavior in order to derive a link between the level of centralization of coercive power

( ) and equilibrium actions following a shock that affects the utility which some citizens receive from con-

forming to their bliss point. I denote equilibrium actions with a superscript * and employ the following

de�nitions:

De�nition 1 A Nash equilibrium is stable in period t if a�j;t = a�j;t�1 8 j, i�1;t = i�1;t�1, and i�2;t = i�2;t�1.

De�nition 2 A shock occurs between periods 0 and 1, where a portion � of the citizenry has the utility they

derive from choosing actions close to their bliss point multiplied by � > 1.12

Assume that a stable equilibrium exists in period zero and that a�0 < i�1;0 and a�0 < i�2;0.13 Assume without

loss of generality that these actions arise from the relationship between the bliss points of the players, where

1N

Pj a

Bj < iB1 < iB2 . That is, the bliss point of the central authority is further away from the mean bliss

point of the citizenry than is the bliss point of the non-central authority. The setup entails that the shock

encourages some of the citizens to transgress the authorities' actions (instead of the less interesting case in

which shocks encourage citizens to choose actions closer to those of the authorities).

The shock occurs in between periods zero and one. The shock encourages some citizens to choose actions

closer to their bliss points, which in turn changes the social norm. The endogenous change in the social norm

encourages more citizens to change their actions in following periods. The economy eventually reaches a

new stable equilibrium in some period t�, which is de�ned as the �rst period in which a stable equilibrium

emerges.

12This de�nition of shock is overly strict in order to allow for a more straight-forward analysis. For example, similar qualitative results

emerge if the shock affects citizens differently.

13Existence is straight-forward to prove, but the proof is omitted here for the sake of conciseness. An existence proof is available upon

request.

7

I begin the analysis by studying equilibrium actions in a stable equilibrium in period 0. Understanding

the actions taken in this period is essential for understanding the effects of the shock in relation to the level

of centralization ( ). Do citizens initially choose actions close to their bliss point? Do the institutional

authorities choose actions close to their bliss points or do they choose actions close to the average action of

the citizens? More importantly, how do the answers to these questions change over different values of the

centralization parameter ( ) and what do these differences entail regarding the relationship between the effect

of the shock and the level of centralization of coercive power?

The �rst order conditions for each of the players (in period t) give the following conditions:

a�j;t =baBj + ci1;t�1 + di2;t�1 + eat�1

b+ c+ d+ e; (4)

i�1;t =fiB1 + gi2;t�1 + hat

f + g + h; (5)

i�2;t =jiB2 + katj + k

: (6)

The steady-state solution in period 0 is derived by employing the equation a0 = 1N

Pj a

�j;0 and the steady

state condition at = at�1. This gives the following:

a0 =1

N

Xj

a�j;0 =ci1;0 + di2;0 + ea0 +

bN

Pj a

Bj

b+ c+ d+ e

=) a�0 =

cfiB1 +c gjiB2j+k

f+ g+h +djiB2j+k +

bN

Pj a

Bj

b+cf+ c gj

j+k

f+ g+h +djj+k

: (7)

Next, consider the interactions between the players. Institutional laws and social pressures encourage

most citizens to choose actions different than their bliss point. Indeed, the two types of pressures interact

with each other. When institutional sanctions are severe (c and d are large), this encourages citizens to choose

actions in favor of the authorities instead of their bliss points, which in turn pushes the social norm closer to

the authority's actions. The latter reinforces the former, as social pressures induce citizens to want to appear

to optimize by choosing certain actions even though they would choose actions closer to their bliss point in

the absence of social pressures. Since citizens only see the actions, not the bliss points, of other citizens,

even a small portion of citizens choosing actions close to the dictates of the authorities can encourage other

citizens to choose such actions, because they are concerned with deviations from the norm a�t . This pushes

8

the norm away from citizens' bliss points, which further encourages actions that deviate from bliss points.

This phenomenon reinforces itself, leading to equilibria, under some parameter sets, where the entire society

chooses actions close to those of the authorities instead of near their own bliss points, even when citizens do

not derive much utility from following the authorities (c and d are small) but derive signi�cant disutility from

transgressing the social norm (e is large). This entails that equilibria can be fragile: it may take only a few

citizens changing actions to undermine the process described above. When some agents change their actions

after the shock and choose actions close to their bliss points, other citizens will be encouraged to follow suit,

and a bandwagon effect ensues.

This process where individuals publicly act in a way that con�icts with their private preferences is related

to preference falsi�cation, a phenomenon that has been studied in great detail by Timur Kuran (1987, 1989,

1995a, 1995b). Kuran (1995a, p. 3) de�nes preference falsi�cation as "the act of misrepresenting one's gen-

uine wants under perceived social pressures". In the context of the model, preference falsi�cation encourages

the bandwagon effect described above. When citizens falsify their preferences, they choose actions which

differ from their bliss point. Part of the reason they do so is that perceived social pressures encourage them

to choose actions close to the social norm - an outcome that is exacerbated when institutional sanctions are

severe (c and d are large). Thus, when a shock causes some citizens to move closer to their bliss points, there

is more for others to gain on the margin from drastically changing their actions to ones closer to their bliss

points, as the social norm moves in this direction. The incentive to change actions in such a manner is greater

where preference falsi�cation is greater, since the marginal bene�t of choosing actions closer to one's bliss

point is greater. Thus, a bandwagon ensues when citizens suf�ciently falsify their preferences.

Yet, despite the importance of preference falsi�cation on equilibrium outcomes, the literature gives no

comprehensive account for the conditions under which it is likely to exist.14 In particular, do certain types of

institutional structures encourage preference falsi�cation? Is there any link between the degree of institutional

centralization ( ) and preference falsi�cation?

There are many means through which preference falsi�cation may be manifested. For example, when

social pressures are signi�cant, it is possible that the social norm will differ signi�cantly from the actions

of the authorities and the citizens. That is, an anarchic focal point could emerge in which citizens' true

14For example, Kuran (1989) emphasizes that revolutionary activity is likely to occur when there exists signi�cant preference falsi�cation

in which citizens publicly support a regime but privately despise it, but he does not specify under which conditions preference falsi�cation

arises.

9

preferences lies somewhere between those of the authorities and their actions. For the remainder of this

paper, however, I will concentrate on the types of equilibria that are more common empirically - namely,

those in which citizens falsify their preferences in favor of the institution. For example, this could account

for individuals falsely declaring loyalty to a government, uncomfortably adopting a hideous fashion trend, or

grudgingly abiding by a religious dictate.

Consider the relationship between centralization ( ) and the degree of preference falsi�cation in period 0.

In highly centralized economies, authority I1 chooses actions closer to the centralized authority, I2, all else

being equal. This encourages the citizenry to choose actions closer to I2, in turn increasing the difference

between the citizens' actions and their bliss points and hence increasing the degree of preference falsi�cation

in the economy.15

Having established a theoretical link between preference falsi�cation and centralization, consider the

effect of the shock on the actions of the citizens in the subsequent stable equilibrium (in period t�). The

utility of citizens affected by the shock is:

USj;t = ��b�aj;t � aBj

�2 � c (aj;t � i1;t�1)2 � d (aj;t � i2;t�1)2 � e (aj;t � at�1)2 : (8)

Thus, the equilibrium action in period t� (and all periods hence) for those affected by the shock is:

a�j;t� =�baBj + ci1;t� + di2;t� + eat�

�b+ c+ d+ e: (9)

To solve for the steady state solution in period t�, employ the equation

at� =1

N

0@Xj;S

a�j;t� +Xj;NS

a�j;t�

1A ; (10)

whereP

j;S a�j;t� is the sum of the actions of the � portion of the population affected by the shock andP

j;NS a�j;t� is the sum of the actions of the 1 � � portion of the population not affected by the shock.

Denoting = b+ c+ d+ e and � = b (� � 1), this entails that

at� =� (ci1;t� + di2;t� + eat�) +

�bN

Pj;S a

Bj

+�+(1� �) (ci1;t� + di2;t� + eat�) + b

N

Pj;NS a

Bj

15This logic is formalized in a proof that is available upon request.

10

=) at� =

( + (1� �)�) cfiB1 +

c gjiB2j+k

f+ g+h +djiB2j+k

!+ �b

N

Pj;S a

Bj +

(+�)bN

Pj;NS a

Bj

��+ ( + (1� �)�)�b+

cf+ c gjj+k

f+ g+h +djj+k

� : (11)

For the rest of the analysis, assume that the average bliss point of a citizen who receives a shock is lower

than the average of those who do not, or 1�N

Pj;S a

Bj <

1(1��)N

Pj;NS a

Bj . This merely entails that those

who receive a shock are on average further away from the actions of the centralized authority.16 I also assume

for simplicity that adding one more citizen to the group who receives a shock does not change the average

bliss point of this group. In other words, 1�N

Pj;S a

Bj is constant. Finally, I assume that all citizens place

suf�cient weight on the costs associated with breaking the dictates of the authorities that these costs are a

salient part of their decision-making calculus.

De�ne the effect of the shock as the difference in mean stable equilibrium actions: � = at� � a0. That

is,� is a measure of the change brought about by the shock. The central question of this paper is, "how does

the institutional centralization of coercive power ( ) affect the degree of change (�) brought upon by some

shock?"

There are there possible answers to this question, depending on the parameters in question. The different

parts of the parameter space which give rise to different equilibrium outcomes are delineated below.

Case 1 The bliss point of the central authority, iB2 , differs greatly from that of the non-central authority, iB1 .

In this case, an increase in centralization ( ) always entails a greater change in equilibrium outcomes

(�), regardless of the size of the shock. This is because the non-central authority, I1, is encouraged to

choose actions close to those chosen by the central authority, I2, whose preferences differ greatly from those

of the citizenry. Thus, citizens choose actions further away from their bliss points, all else being equal,

entailing that when citizens with the most to gain decrease their actions after the shock, others have greater

incentive to also decrease their actions, since their marginal return is larger. These actions encourage both

authorities to decrease their actions, and this process continues until the old equilibrium unravels and a new

stable equilibrium emerges. This outcome - a shock leading to a substantial change in equilibria - is similar

16Removing this assumption affects some equilibrium outcomes, but does not affect the fundamental intuition, and is thus employed for

simplicity. Indeed, to the extent that the authorities can steer the shock to certain portions of the population, those that choose actions

differing from the wishes of the authorities are likely those whom they would choose, so as to prevent a critical mass of opponents from

forming.

11

to outcomes in the herding and informational cascades literature, but for fundamentally different reasons.

Those works suggest that signi�cant changes in equilibrium outcomes occur because individuals cease using

private information and instead follow common-knowledge signals (Banerjee 1992, 1993; Bikhchandani,

Hirshleifer, and Welch 1992, 1998). In this model, which is closer to Kuran (1987, 1989, 1995a, 1995b,

1998) and Bernheim (1994), shocks encourage actions which change social norms to a suf�cient degree that

cascades occur despite citizens taking into account their private information or types.17

Case 2 The bliss point of the central authority, iB2 , does not differ greatly from that of the non-central au-

thority, iB1 , and the shock is not systemic (� is small).

In this case, a cascade never emerges and centralized authorities are better able to suppress dissent -

even when the magnitude of the shock, �, is large. This occurs when � is small enough that the change in

actions by the citizens affected by the shock does not suf�ciently change the norm so that a cascade forms.

Moreover, unlike in Case 1, citizens do not falsify their preferences to as signi�cant of a degree, since the

central authority chooses actions closer to the mean action of the citizenry when iB2 is small. This further

decreases the size of the parameter set over which a cascade emerges.

Case 3 The bliss point of the central authority, iB2 , does not differ greatly from that of the non-central au-

thority, iB1 , and the shock is systemic (� is large).

In this case, the level of change (�) is increasing in the degree of centralization only when the shock size

(�) is suf�ciently large. The intuition underlying this result is that if the shock is suf�ciently small, the social

norm does not change dramatically and so a cascade never forms. In the absence of a cascade, decentralized

institutional authorities (I1) are more responsive to changes in the mean action of the citizenry, as they face

lower costs from differing from the dictates of the central authority for responding. On the other hand, a

suf�ciently large shock encourages a cascade, and the process described in Case 1 follows whereby the level

of social change (�) is increasing in the degree of centralization ( ).

The intuition underlying these three cases is formalized in Proposition 1. The proof is in Appendix A.

17This concept is also similar to the concept of "availability cascades" analyzed by Kuran and Sunstein (1999), in which expressed

perceptions help form the perceptions of others, which in turn makes these perceptions increasingly likely through their availability in

public discourse. Unlike this work, however, I am not concerned with information revelation (such as expanding the scope of public

discourse) beyond that of one's type.

12

Proposition 1 There are three possible outcomes relating the level of change (�) and the level of centraliza-

tion ( ). In the � � iB2 plane, the parameter set over which each case emerges is demonstrated graphically

in Figure 1. These outcomes are:

Case 1)� is increasing in . This occurs when iB2 is suf�ciently large.

Case 2)� is decreasing in . This occurs when iB2 and � are suf�ciently small.

Case 3) 9 some �� such that � is decreasing in when � < �� and � is increasing in when

� > ��. This occurs when iB2 is suf�ciently small and � is suf�ciently large.

Moreover, the threshold level �� in Case 3 is decreasing in � and iB2 .

[FIGURE 1 HERE]

Case 3 is shown graphically in Figure 2. This �gure shows the effect of the shock (�) on the the level of

social change (�) for economies with two different initial levels of institutional centralization ( H and L,

where H > L).

[FIGURE 2 HERE]

2.3 Discussion

Proposition 1 formalizes numerous aspects of the relationship between the centralization of coercive power,

preferences, and massive equilibrium change. Yet it also begs as many questions as it answers. Such ques-

tions include, but are not limited to: "Under what real world conditions are the three cases espoused above

realized?" "What type of actions are represented in each of the cases?"

Case 1 emerges when the desires (bliss points) of the centralized authority are not aligned with those

of the other authorities and the citizenry. Taking such actions thus bene�ts the authority at the expense of

everyone else. Hence, Case 1 does not represent most types of economic actions - although the authority

may have different distributional desires than the citizens, both groups can be better off when ef�ciency

is maximized.18 If the action in question is an economic one, such as taxation, upholding property rights,

providing social services, or printing money, the bliss point of the central authority will diverge from the

citizenry, but likely not by enough for Case to 1 emerge. On the other hand, certain types of political or

18Case 1 could represent an economic action if the "centralized authority" is a group of powerful special interests acting collectively. Such

special interests often have the ability to dole out political and economic sanctions - sometimes through extra-legal means - and thus

could play the role that the central authority plays in the model.

13

social actions may be represented in Case 1, especially actions that are designed to keep the central authority

in power. Examples of these types of actions include suppression of free speech, religion, or the media,

election rigging, or the formation of secret police forces. The type of shocks which may be associated with

such examples include the public outing of brutal state tactics, an obviously rigged election, or the capturing

of a popular opposition �gure. In any of these examples, Case 1 predicts that increased centralization of

coercive power always leads to greater change after a shock occurs, as the citizenry was privately opposed to

the authority's actions before the shock, and thus a cascade is more likely to arise.

In Cases 2 and 3, however, the bliss point of the centralized authority does not differ substantially from

those of the other authorities and the citizenry. In other words, the actions in question are the those in

which the preferences of the institutional authorities are broadly aligned with those of the citizenry and both

can bene�t from their implementation. Such characteristics are often found in economic policies, such as

subsidies on essential goods, trade restrictions, or setting interest rates. Case 2 arises when shocks are not

systemic; this may occur, for example, when an economic shock affects one industry. In this case, the shock

affects a small enough portion of the population that no cascade emerges, and centralized authorities are

better able to suppress dissent. On the other hand, Case 3 arises when shocks are systemic. Such shocks

occur when macroeconomic variables such as in�ation, interest rates, or government debt suddenly change.

In such a case, centralized authorities are better able to suppress dissent when the shock level is small (for

example, when in�ation rises by a small amount), but are more susceptible to massive changes when shock

levels are large (for example, in periods of hyperin�ation).

The three cases detailed in Proposition 1 thus represent different types of shocks. A qualitative represen-

tation is shown in Figure 3.

[FIGURE 3 HERE]

Since this paper is concerned with massive economic, political, and social change, I concentrate on Cases

1 and 3. The intuition espoused above offers two explanations for why apparently tranquil, centralized soci-

eties can quickly undergo massive changes. One explanation, which is also offered by Kuran (1989, 1995a,

1995b), is that preference falsi�cation can encourage a latent bandwagon to unfold after a shock. That is,

an economic, political, or social shock may move the equilibrium social norm by enough to encourage all

citizens, even those who are not directly affected by the shock, to choose drastically different actions.

The other explanation, which is novel to this paper, sheds light on the potential effects of economic shocks

(Case 3). It suggests that a high degree of centralization of coercive power discourages marginal changes to

14

equilibrium actions after small economic shocks. Citizens have less incentive to transgress the dictates of the

authorities, as they incur two institutional costs from doing so, and fear of multi-lateral punishment discour-

ages citizens from "pushing the envelope" after a small shock. The same citizens may be more encouraged

to "push the envelope" in less centralized societies, however, as they face less cost from doing so and the

authorities (especially I1) react by changing the actions to a greater extent. On the other hand, when larger

economic shocks materialize, greater equilibrium changes are more likely to result in highly centralized

economies. This occurs because citizens falsify their preferences to a greater extent in such economies, and

thus large shocks encourage some citizens to change their actions, in turn making it more likely that a cascade

away from the authority's dictate will arise.

Kuran (1995a, 1995b) does suggest that unanticipated regime change is more likely to occur in politically

repressive countries. His hypothesis coincides to some degree with the one made in the present paper, though

it is not clear that Kuran's hypothesis holds when other types of freedoms (religious, economic, legal) exist

in politically repressive regimes. The essential difference between the present hypothesis and Kuran's is that

I stress the interdependence of institutions that are able to impose different types of sanctions. This leads to a

similar conclusion as Kuran, as such institutional structures are often found in politically repressive regimes.

This analysis also helps underscore an important feature underlying political legitimacy: legitimacy exists

not only when people believe in the legitimacy of the government, but when they believe that other people

will act in a manner that supports the legitimacy of the government. This is in line with Greif (2006, p. 138-

40), who argues that a set of institutionalized rules (in this case, the legitimacy of the government) is con�ned

to those that are reproduced, and not refuted, by their implied behaviors. That is, equilibrium actions require

that individuals correctly anticipate each others actions regardless of the beliefs underlying these actions. In

turn, this creates a self-enforcing belief in the beliefs of others. There is a subtle but important difference

between this notion of legitimacy and a more conventional, Weberian notion espoused by Razi (1987, 1990),

who views legitimacy as being dependent on "the extent to which the relevant portion of the population

perceives that the regime is behaving [legitimately]" according to an established set of norms.19 Instead, I

suggest (in line with Kuran [1995]) that when a centralized authority acts "illegitimately" (in terms of the

model, it increases i2), it only loses legitimacy when citizens see others acting in a way that undermines

legitimacy. Otherwise, citizens may continue to falsify their preferences, and the government will still be

19See Greif (2010) for a similar de�nition.

15

viewed as legitimate.

The model also sheds light on the connection between institutional centralization and revolutions. Numer-

ous de�nitions for revolution exist in scholarly works.20 For example, Goldstone (2001) de�nes revolution

as "an effort to transform the political institutions and the justi�cations for political authority in a society, ac-

companied by formal or informal mass mobilization and non-institutionalized actions that undermine existing

authorities." Kuran (1989, 1995a) broadly de�nes a revolution as a discontinuous change in public opinion or

social order and Davies (1962) de�nes revolutions as "violent civil disturbances that cause the displacement

of one ruling group by another that has a broader popular basis for support."

The model suggests that revolutions may occur after an economic shock only when the shock is suf�-

ciently large; on the other hand, smaller shocks are more easily repressed in centralized regimes. "Shock" is

a somewhat ambiguous term, but the literature on revolutions provides numerous types of shocks that could

precipitate revolutions, such as sharp reversals in economic fortunes (Davies 1962; Tanter and Midlarsky

1967), rapid economic growth (Olson 1963), defeat in war, or sustained population growth (Goldstone 2001).

The analysis also indicates that social pressures may assist organizations in overcoming free-riding prob-

lems potentially associated with revolutionary activity. Tullock (1971) argues that revolutions are public

goods, and that the individually-rational action is to never participate in revolutions. Tullock's model, how-

ever, fails to incorporate utility derived from social interaction.21 The present model derives a starkly differ-

ent result largely due to the incorporation of social costs, suggesting that once a cascade begins, social costs

avoided by joining the revolution may be greater than the costs (that are avoided by free-riders) imposed by

one of the authorities.22

Finally, consider the implications of the model for centralized regimes. Holding the degree of centraliza-

20There is a large literature merely attempting to de�ne the term revolution. I have no desire to enter this debate, summarized nicely by

Goldstone (2001), and instead note that revolutions are an extreme form of massive equilibrium change.

21On the other hand, Olson (1971) recognizes that social pressures may work to overcome free-riding problems in small groups, but these

sanctions cannot work in larger groups. As Kuran (1995a) points out, Olson's analysis has dif�culty accounting for collective action

when public expression of preferences is costly.

22This intuition is supported by Goldstone (2001), who argues that, unlike Tullock (1971), revolutionary actors do not act, or even think

of themselves, as acting alone. Instead, social ties - established by preexisting networks - play an instrumental role in recruitment. In a

similar vein, Goldstone (1994) suggests that the group, not the individual, is the relevant unit of analysis when studying revolution. This

insight is not necessarily contradictory to the one presented in the present paper. Indeed, in many cases the existence of such groups is

the reason that social costs and cascades originate.

16

tion ( ) constant, which parameters affect the likelihood of revolution in such regimes? Proposition 1 suggests

that �� (the shock level at which equilibrium change is greater in the centralized regime for all � > ��) is

decreasing in iB2 and �. Thus, a straight-forward way for centralized authorities to prevent revolutions is to

attempt to decrease �. This also pushes the economy closer to Case 2, where centralized authorities can more

easily suppress dissent. This often takes the form of the authority attempting to do everything possible to

ensure that if an economic shock occurs, it is spread to as small of a portion of the population as possible.

This provides a reason why centralized, authoritarian governments often impose restrictions on domestic and

international media. Public dissatisfaction expressed in protests or other types of grievances is more likely to

result in a bandwagon the more that people know that others are willing to act in a way that coincides with

their own intrinsic beliefs. Hence, media suppression is more likely to exist in economies where coercive

power is centralized, since anti-government bandwagons of public opinion are more likely to form as a result

of a systemic shock in such economies.

The following testable predictions arise from the model:

� Economies where coercive power is centralized are more insulated from change resulting from small

economic shocks (small �) than decentralized economies. However, massive changes in equilibrium

outcomes are more likely to emerge in centralized economies, as bandwagons are more likely to form

upon a larger economic shock (large �).

� Centralized authorities are more likely to suppress dissent if shocks are not spread to much of the

population (small �), while decentralized authorities are more likely to marginally change their actions

in accordance with changing social norms. On the other hand, centralized authorities will not be able

to suppress dissent when shocks are felt by larger portions of the population (large �), and massive

equilibrium changes will result.

� Revolutions or massive equilibrium changes are more likely to occur in centralized economies regard-

less of the size of the shock when the preferences of the authority (iB2 ) diverges greatly from those of

the citizenry (aBj ). This may occur when the shock is political or social in nature; examples include

governments that represent the interests of a small group of oligarchs or corporations at the expense of

the mass of the population.

17

3 Austerity Riots and Centralization

A primary implication of the model is that countries with centralized institutions are more insulated from

change when economic shocks are small but are more susceptible to sudden, massive changes when economic

shocks are large (Case 3). In this section, I test this hypothesis by analyzing the severity of austerity protests

in the developing world between 1976 and 1992. Such protests were common in the developing world

beginning in the mid-1970s in reaction to measures employed - almost always as a condition of IMF aid - to

combat in�ation and government debt. I test the relationship between a series of "IMF pressure" variables

and severity of protests over differing degrees of institutional centralization to shed light on the relationships

espoused in the model. Although the data cannot speak to the microeconomic mechanisms highlighted in

the model (namely, those related to internal and expressed preferences), it can shed light on the connections

between macroeconomic events and changes in publicly expressed opinions.

Modern austerity protests began in the mid-1970s, with the �rst one occurring in Peru in 1976. The

protests were sparked by austerity measures which were almost always imposed by the IMF as a condition of

assistance. The stated aims of these measures were freeing up markets and cutting government spending in

order to reduce government debt and curb massive in�ation. These market-based measures, known by some

as �shock treatment�, included currency devaluation, broad reduction of spending on the public sector, priva-

tization of state-owned corporations, cuts in public subsidies for food and basic necessities, wage restraints,

higher interest rates, and elimination of protectionism (Walton and Seddon 1994).23

These policies sparked protests in many places where they were imposed. Such protests were de�ned

by Walton and Seddon (1994) as �large scale collective actions including political demonstrations, general

strikes, and riots, which are animated by grievances over state policies of economic liberalization imple-

mented in response to the debt crisis and market reforms urged by international agencies.� The international

agency most associated with these protests is the IMF, and hence Joseph Stiglitz has named them �IMF riots�.

The distributional implications of these policies are clear � most policies negatively affected the urban

poor, at least in the short run (Walton and Ragin 1990; Walton and Seddon 1994). The protests were primarily

urban in nature, often following a rise in a price for a speci�c good or an elimination of a subsidy. In some

cases, the result was relegated to one city and remained non-violent, such as organized strikes planned in

Ecuador and Bolivia (Walton and Ragin 1990). On the other extreme, protests turned into deadly riots which

23For a scathing review of these policies over the last half-century, see Klein (2007).

18

spread throughout the country, as was the case of the Venezuelan protest of 1989, where a week of rioting

spread from Caracas to 16 other cities.

What determines the differences in severity of these protests? This topic has received some attention

from sociologists, who have proposed a wide range of explanations. Walton and Seddon (1994) and Walton

and Ragin (1990) provide statistical evidence that overurbanization plays a key role in both the presence and

severity of the riots. They suggest that the linkage between the two lies in the development of organizational

infrastructure capable of mobilizing political action. Walton and Ragin (1990) also suggest that IMF pressure

(which they proxy for with four different indicators to create an index of IMF involvement) signi�cantly

affects protest severity but in�ation and debt do not.24

In this section, I analyze how changes in IMF involvement affect the likelihood of severe protest in a coun-

try. When IMF pressure is suf�cient, pressure to liberalize markets is likely to ensue, and protests may follow.

The model suggests that when IMF pressure (the "shock" in the model) is small, centralized economies will

better be able to suppress protests, but marginal changes will occur in decentralized economies. However,

signi�cant IMF pressure is more likely to precipitate massive changes in centralized economies. In other

words, the following prediction arises from the model:

Prediction 1: When there is a small amount of IMF pressure, austerity protests will on average be less

severe in countries with more centralized institutions, all else being equal. However, when there is a large

amount of IMF pressure, austerity protests will on average be more severe in countries with more centralized

institutions.

3.1 Data

In order to limit statistical dif�culties arising from analyzing a biased subset of riots, I have gathered data

on austerity protests covering the same years as Walton and Seddon (1994): 1976-1992. The former date

denotes the onset of the �rst modern austerity protest while the latter represents the time in which Walton and

Seddon went to press.

As noted by Walton and Ragin (1990), obtaining a complete list of debtor countries that experienced

international pressure to implement austerity measures can only be formulated indirectly, as the exact terms

24On the other hand, Auvinen (1997) �nds that poor economic performance (indicated by high in�ation and large debt service) is associated

with political demonstrations, riots, and strikes.

19

negotiated between the IMF and debtor countries is kept secret. To this end, I employ three measures identi-

�ed by Walter and Ragin as indicative of IMF pressure: 1) the country employed the "extended fund facility"

(EFF), generally reserved for countries suffering a signi�cant imbalance of payments relating to structural

maladjustments in production and trade (IMF [various]), in a given year between 1976 and 1992;25 2) the

country's ratio of IMF funds used to its IMF quota exceeded 125% in a given year between 1976 and 1992;

3) the country rescheduled or renegotiated its debt in a given year between 1976 and 1992.26 70 countries

satis�ed one of these three criteria between 1976 and 1992.27 I form a panel which is restricted to years in

which one of these criteria were satis�ed in the country in question within one year (either before or after).

A list of these countries and the years employed in the data is listed in Table 1.

[TABLE 1 HERE]

40 countries in the data set experienced austerity protests between 1976 and 1992. A Lexis-Nexis search

of news reports of the 70 countries listed in Table 1 (as well as any listed in Walton and Seddon [1994]

that were not in Table 1) from 1976-1992 produced 116 separate instances of austerity protest. Protests and

riots were only documented if they resulted from austerity measures or IMF pressure - other types of anti-

government protests or strikes are not included in the data. Planned general strikes and small, non-violent,

sector-speci�c strikes are also not included in the data (even if they resulted from austerity measures), as this

paper is intended to investigate shocks that affect the population as a whole (high �).28

25The IMF de�nes the extended fund facility as "An IMF lending facility established in 1974 to assist member countries in overcoming

balance of payments problems that stem largely from structural problems and require a longer period of adjustment than is possible under

a Stand-By Arrangement. A member requesting an Extended Arrangement outlines its objectives and policies for the whole period of

the arrangement (typically three years) and presents a detailed statement each year of the policies and measures it plans to pursue over

the next 12 months." (IMF)

26Walter and Ragin split the last category into two: debt rescheduling and debt renegotiation. My reading of the data suggests that the

line between these two is often blurred, so I have lumped them together. Data for EFF and IMF quota comes from various IMF Annual

Reports; data for IMF funds used comes from the World Development Indicators database; data for debt rescheduling and restructuring

comes from clubdeparis.org, Kuhn and Guzmàn (1990), and Dillon et al. (1985).

27Some countries were omitted from the dataset due to lack of IMF or control data. These include:Angola, Barbados, Burma, Dominica,

Equatorial Guinea, Grenada, Liberia, Somalia, Uganda, Western Samoa, Yemen, and Yugoslavia. Iran and South Korea did not satisfy

any of these conditions but are included in the data since there was an austerity protest in each country.

28All results are broadly robust to inclusion of general strikes and industry-speci�c protests as severity 1 protests. These results are not

included but are available upon request.

20

I subjectively coded each of these protests by severity using the following criterion. An instance was

scored 1 if it were a small, con�ned (to one or two cities) protest or general strike, 2 if it were either pro-

longed but con�ned or widespread but not prolonged, with minimal deaths resulting, and 3 if the protest

were prolonged, contained widespread riots, and resulted in many deaths. Only protests carrying all three

of these characteristics were coded level 3. An example of a protest coded 3 occurred in Algeria in October

1988, when 159 protesters were killed in a few major Algerian cities over the span of week and thousands

were injured and arrested. While coding these protests is an admittedly subjective process, a reading of the

articles reporting on the protests generally showed that most protests/riots easily �t into one of these three

categories. The differences between protests coded 1 and other protests are especially stark. The difference

between protests coded 2 and 3 are less obvious and thus more subjective, but I control for this in the analysis

by lumping these two types of protests together. The interested reader can reference Appendix D, which

provides a description of how the protest data were collected as well as brief summary of the salient features

of each of the 116 protests. I also create an alternative index, noted in this table, for protests in which the

severity level was not obvious. All results are robust to this alternative speci�cation. Tables 2 through 4 and

Figure 4 show the data and summary statistics of these protests by country, year, and continent.

[TABLES 2, 3, AND 4 AND FIGURE 4 HERE]

Numerous proxies are available for the degree of �centralization� of coercive power. The intuition out-

lined in the model indicates that a good proxy is one that accounts for one authority's ability to affect numer-

ous types of sanctions. One such variable is spelled out in the Polity IV data set (Marshall and Jaggers 2008):

constraint on the executive. This variable is de�ned as:

The extent of institutionalized constraints on the decision-making powers of chief executives,

whether individuals or collectivities. Such limitations may be imposed by any "accountability

groups." In Western democracies these are usually legislatures. Other kinds of accountability

groups are the ruling party in a one-party state; councils of nobles or powerful advisors in monar-

chies; the military in coup-prone polities; and in many states a strong, independent judiciary.

This variable ranges from 1 to 7, with 1 equaling �Unlimited Authority� (no regular limitations on the

executive's actions) and 7 equaling �Executive Parity or Subordination� (accountability groups have effective

authority equal to or greater than the executive in most areas of activity). This variable provides an ideal proxy

21

for the degree of institutional centralization, as spelled out in the model, because it measures the degree to

which political authorities can extend multifarious sanctions, especially judicial ones.

Another mechanism through which the centralization of coercive power provides the conditions for

protest is that more centralized regimes are better able to suppress individual freedoms - in particular, po-

litical freedom. In this light, another, less direct proxy for the "centralization" of coercive power is a degree

of political freedom variable provided by Freedom House (2009). This variable ranges from 1 to 7, with 1

equaling "most free" and 7 equaling "least free". Although this is a less direct measure of institutional cen-

tralization than the constraint on the executive variable, the model suggests that similar patterns should be

seen when either of these variables are employed as a proxy for centralization.

Other controls are employed to account for phenomena which economists and sociologists consider as

salient factors associated with protest activity. These include the urbanization rate, per capita GDP, popula-

tion, and a measure of religious fractionalization.29 The religious fractionalization index is constructed like a

Her�ndahl index, equaling the sum of the proportion of each religion in the country squared. The summary

statistics of all controls are reported in Table 5.

[TABLE 5 HERE]

3.2 Analysis

3.2.1 Testing for the Presence of Protest

The data provide a chance to test the model's primary prediction: countries with centralized coercive author-

ities should have smaller changes in public opinion when shocks are small, but larger changes when shocks

are large. Yet, before we can test how shocks (proxied with IMF variables) affect the severity of protests, we

must establish whether the proxies are predictors of protest. In particular, the model suggests the testing of

the following equation, where Protest is a dummy equaling 1 if there is a protest in a given year and country,

29Data on urbanization, GDP, and population comes from World Bank (various). The measure of religious fractionalization is derived

from data found in Barrett, Kurian, and Johnson (2001).

22

i is the country, t is the year, and X is the set of control variables:30

Protestit = �0+�1Shockit+�2Centralization Proxyit+�3Shockit * Centralization Proxyit+�Xit+"it (12)

There are 3 different IMF variables that proxy for a shock: use of IMF funds (divided by IMF quota),

use of EFF funds (divided by IMF quota), and the number of restructurings and reschedulings. The �rst two

variables are de�ated by the country's IMF quota, which is based on its standing in the world's economy.

Walton and Ragin (1990) note that all three of these variables are good predictors of austerity protests, with

the last one being the strongest predictor.31 I also create variables for the greatest single-year level of the

three IMF proxies over the previous two years. These latter variables may be more suitable, since the chain of

events from IMF pressure to implementation of austerity programs to protests can take months to matriculate

and are thus sometimes realized over multiple years.

For the centralization proxy, the Polity IV "constraint on the executive" and the Freedom House "Political

Freedom" variables are available. Since these variables range from 1-7, it is not clear exactly how to interpret

the regression coef�cients. For example, is the difference between 1 and 2 the same as the difference between

5 and 6? To overcome this problem, I create three dummy variables: i) equals one if the constraint on the

executive variable is 1 (least constraint) and 0 if it equals 2-7; ii) equals one if the constraint on the executive

variable is 1 or 2 and 0 if it equals 3-7; iii) equals one if the political freedom variable is 6 or 7 (least free)

and 0 if it equals 1-5.32 In all of these variables, the centralization proxy is greater when the dummy equals

one. Robustness checks using different cutoff points for the dummy variables (e.g. equaling one if constraint

30The controls employed in these regressions are broadly consistent with those pointed out by Auvinen (1997) as ones that traditionally

have received attention in the economic and sociology literature related to political con�ict. They include urbanization rate, real per

capita GDP, population, religious fractionalization, and the number of previous protests. In�ation and government debt are not included

as controls because they frequently lead to IMF involvement.

31Walton and Ragin formulate an "IMF pressure index" which is the summation of the z-scores of all 4 IMF indicators. I do not do this

for two reasons: 1) I �nd, like Walton and Ragin, that the number of restructurings and reschedulings is by far the best predictor; 2) the

sum of the z-scores is dominated by the restructuring variable, which is not normally distributed, and hence converting it to a z-score is

erroneous. I have analyzed the regressions reported in this paper using Walton and Ragin's z-score variable and the results are similar -

though not as strong - as the ones reported here. These results are available upon request.

32I also created a dummy equaling one if the political freedom variable 7 (least free) and 0 if it equals 1-6, but many of the regressions had

missing values and I have not included these results, which are available upon request.

23

is less than 4) provide weaker but broadly similar results. These are not reported because their interpretation

is much less straight-forward - the model makes clear predictions about the differences between severely

centralized authorities and not severely centralized authorities, but there may be ambiguities (especially if

the relationship between centralization and protests is non-linear) in the differences between moderately

centralized authorities and other authorities.

Equation (12) does not provide a test of the model's hypothesis - instead, it provides a test of how well

the IMF variables proxy for a "shock" in the context of the model. The model indicates that the coef�cient on

the IMF "shock" variable, �1, should be positive (and statistically signi�cant). If it is not, then the variable

in question is not a predictor of protests in general, and thus should shed little light on how institutional

centralization affects the severity of protests.

Yet, not all "shocks" are alike - some are weak (i.e., leading to little austerity) and some are strong. Thus,

the interaction between the shock and the centralization proxy can also tell us something about the strength

of the shock. For one, we would expect a "strong" shock to increase the probability of protest regardless of

the degree of centralization. Hence, the coef�cient on the interaction term, �3, will be small and insigni�cant

in regressions using "strong" IMF variables. However, if the shock is "weak" - that is, one that leads to few

or small austerity measures being taken - the intuition of the model suggests that more centralized economies

will experience smaller shocks, since authorities are better able to suppress dissent. That is, the coef�cient on

the interaction term, �3, will be negative and signi�cant in regressions using "weak" IMF variables. In other

words, the regression model in equation (12) provides us with the following insights:

1. An IMF pressure variable is not a good predictor of protests if �1 is statistically insigni�cant. If �1 is

statistically insigni�cant, then:

2. An IMF pressure variable is a "weak" predictor of protests if �3 is negative and statistically signi�cant

3. An IMF pressure variable is a "strong" predictor of protests if �3 is statistically insigni�cant

The above delineation of weak versus strong proxies is not meant to support any prediction of the model.

Indeed, the model is used to de�ne whether a shock is weak or strong - thus, the "strength" of the IMF

pressure proxy is determined by de�nition, not empirics. However, these de�nitions are useful later for testing

the affects of centralization on protest severity given that a protest occurs. Since protests of all severities are

lumped together in equation (12), the delineation between weak and strong proxies should only be viewed as

one that allows for cleaner testing in the following section of the main prediction of the model.

24

A logit model is used to test this regression. It is possible that some of the years with zero protests are

censored. If international news reports did not pick up on an austerity protest, then it is not included in

the data. However, where this is true it likely works against the hypotheses presented in this paper, as more

centralized governments are generally more able to suppress news coverage of protests. Hence, any censoring

will make the IMF proxies appear weaker than they actually are.33

The regression results are reported in Table 6. The only results reported are the coef�cients on the "Shock

Variable" (�1) and the interaction term (�3). All other coef�cients and regression statistics are not reported,

but are available by request. All regressions also include continent and year dummies.34

[TABLE 6 HERE]

The results in Table 6 indicate that the "use of IMF funds" variable is not a good predictor of protests,

but the "EFF" and "restructuring/rescheduling" variables are good predictors. That is, the coef�cient on the

IMF Pressure Index, �1, is insigni�cant in all but one of the regressions where "use of IMF funds" is the

shock proxy but is always signi�cant when the EFF and restructuring variables are used as proxies. However,

the EFF variable appears to be a "weak" predictor, as in nearly every regression the interaction term, �3, is

negative and signi�cant. As noted above, the interpretation of this result is that an increase in these variables

have a positive effect on the probability of prediction when centralization (or freedom) is suf�ciently small,

but in highly centralized (or unfree) countries, the effect is small or even negative. The model suggests that

this is what we would expect to see when shocks are small - marginal changes may result in decentralized

economies, but centralized authorities will be able to suppress the effects of such a shock.

On the other hand, the number of restructurings or reschedulings appears to be a "strong" predictor of

protests, a result also found in Walton and Ragin (1990). The coef�cient on the proxy (�1) is positive and

highly signi�cant in all regressions, but the coef�cient on the interaction term (�3) is highly insigni�cant in

all regressions. This indicates that an increase in the number of restructurings or reschedulings has a positive

affect on the probability of protest regardless of the degree of centralization. Indeed, the terms of IMF

33Alternatively, I have tested all regressions in this paper using a Tobit model with the lower limit censored at 0. All results are broadly

the same. These results are available upon request. For a particularly relevant discussion on the use of Tobit and interpreting regression

coef�cients, see Roncek (1992).

34Including country dummies would be ideal, but there is too little variation within countries over time for this to be a feasible approach,

as all results are dependent on a small number of observations.

25

"conditionality" (policy prescriptions that are agreed to by all parties) are generally agreed to in return for the

renegotiation or rescheduling of debt repayment (Walton and Ragin 1990), so it should not be surprising that

this variable is the "strongest" predictor of protests.

It is also possible that the EFF and restructuring proxies have independent effects (or close to independent)

on the probability of protest. Indeed, the correlation between the two variables is 0.142. The two variables are

not independent, but it is possible that regressions 6.1-6.3 suffer from omitted variable bias if both of these

proxies are not controlled for in the same regression. Indeed, of the 274 observations (year and country) where

the rescheduling variable is positive, only 47 (17.2%) also have positive EFF variables. On the other hand,

only 47 of the 133 observations where EFF variable is positive (35.3%) also have a positive rescheduling

variable. This entails that not accounting for both of these "shock" proxies in the same regression may lead

to biased results. In other words, a more appropriate regression model is represented by equation (13):

Protestit = �0 + �1AEFFit + �1BReschedulingit + �2Centralization Proxyit+

�3AEFFit * Centralization Proxyit + �3BReschedulingit * Centralization Proxyit + �Xit + "it(13)

Table 7 reports the results of this regression. It appears that both the EFF and Rescheduling proxies

remain good predictors of protests, as �1A and �1B are positive and signi�cant in almost all regressions.

The EFF variable remains in most regressions as a "weak predictor", as �3A is negative in all regressions

and signi�cant at the 11% level in 4 of the 6 regressions. The rescheduling proxy also remains as a "strong

predictor", as �3B is highly insigni�cant in all regressions.

[TABLE 7 HERE]

3.2.2 Relating "Centralization" and Protest Severity

The analysis in the previous section suggested that the EFF variable is a weak predictor of protests while

the number of restructurings or reschedulings is a strong predictor. On the one hand, the terms "weak" and

"strong" are merely artifacts of the model. On the other hand, delineating these variables by their "strength"

allows for the testing of model's primary prediction. This is, namely, that small shocks will result in more

change in decentralized economies, as centralized authorities are more able to suppress small, marginal

changes in actions. However, larger shocks will lead to larger changes in centralized economies. In terms of

the variables in question, the following prediction emerges:

26

Prediction 2: All else being equal, a change to a more centralized economy:

i) has no signi�cant effect (or possibly a slight negative effect) on the probability of a more severe

protest occurring when the EFF and restructuring variables are small (no shock)

ii) has a negative (and possibly signi�cant) effect on the probability of a more severe protest occur-

ring when the EFF variable is large but the restructuring variable is small (weak shock)

iii) has a positive and signi�cant effect on the probability of a more severe protest occurring when

the restructuring variable is large (strong shock)

Prediction 2 suggests that there is a relationship between the "strength" of the shock and the severity of

protests. Thus, an appropriate test should consider the determinants of protest severity given that a protest

occurred. Including non-protests years is akin to comparing apples with oranges; and since the previous

regressions showed that the EFF and rescheduling proxies are good predictors of protest activity, it is not

necessary to include non-protest years in a regression aimed at shedding light on the determinants of protest

severity. Not including non-protest years also has the property that we can test the different predictions

relating to "weak" and "strong" shocks. If non-protest years are included, "weak" and "strong" shocks will

have certain properties arise via de�nition rather than empirics, as these terms are de�ned in the context of

regressions analyzing the determinants of protest activity (0 or 1).

Testing the regression model in equation (14) sheds light on the above prediction. The dependent variable,

Protest Severityit, equals one if the most severe protest in a given year and country equals 2 or 3 and equals

zero if the most severe protest equals one. Unfortunately, there are too few observations of protests of severity

3 to test the model using a dependent variable which equals one only if the protest is of severity 3. This should

not detract from the results, however, as the most signi�cant differences in protest severity are between those

coded 1, which are generally con�ned and short, and 2 and 3, which are more widespread, long, and violent.

Protest Severityit = �0 + �1AEFFit + �1BReschedulingit + �2Centralization Proxyit+

�3AEFFit * Centralization Proxyit + �3BReschedulingit * Centralization Proxyit + �Xit + "it(14)

Prediction 2 suggests that �2 is insigni�cantly different from zero, as this variable can be interpreted as

the effect of increased centralization when both the EFF and rescheduling variables are zero, and thus there

is no or little shock. If anything, this coef�cient should be negative, since protests resulting from unobserved

shocks (that do not show up in the EFF or rescheduling variables) should be more easily suppressed in

27

centralized economies. It also suggests that �3A is negative and possibly signi�cant,35 as it can be interpreted

as the additional effect of increased centralization when there is an increase in the EFF variable (a "weak

shock") but the rescheduling variable is zero. Finally, the model suggests that �3B is positive and signi�cantly

different from zero, as this coef�cient can be interpreted as the additional effect of increased centralization

when there is an increase in the number of reschedulings (a "strong" shock). These predictions are broadly

con�rmed in Table 8, which reports the results of logit regressions represented in equation (14):

[TABLE 8 HERE]

These regressions broadly support the predictions of the model. The coef�cient on the interaction term

of EFF and centralization variables, �3A, is negative in all speci�cations, although it is never statistically

signi�cant. On the other hand, the coef�cient on the interaction term of the rescheduling centralization

variables, �3B , is positive in all regressions and signi�cant in four of them. It is highly signi�cant in all three

regressions where the rescheduling variable is considered over a two year span, which is a more theoretically

appealing metric since the time between the onset of IMF pressure and protest often extends over the span

of one year. Finally, the coef�cient on the centralization proxy term, �2, is highly insigni�cant in four of the

regressions, though it is signi�cant in the other two. As noted above, the negative coef�cient in the bottom

panel of regression 8.2 is consistent with the model, as protests arising from smaller, unobserved shocks

should be more easily suppressed in a centralized economy. On the other hand, the positive and highly

signi�cant coef�cient on �2 in the middle panel of regression 8.1 does not support the model's predictions;

however, this coef�cient is highly insigni�cant when a more appropriate metric - the "shock" over 2 years

(regressions 8.2) - is considered.

Yet, although the sign of the coef�cients broadly accord with the predictions of the model, the coef�cients

are dif�cult to interpret. As Norton, Wang, and Ai (2004) point out, the cross-partial derivative of two

continuous terms that are interacted is not the coef�cient on the interaction term, as it is in OLS. Instead,

we are concerned with the marginal effects, which can be calculated at different levels of "shock" (EFF and

rescheduling). That is, the following equation provides the marginal effects:

35The model does not necessarily indicate that the interaction effect will be signi�cant when the shock is "weak" because a "weak" shock

may not spur massive protests in either centralized or decentralized economies. It does suggest, however, that the point estimate should

be negative.

28

4F (u)4CP = F (u;CP = 1)� F (u;CP = 0) .

where F (u) is the logit cumulative distribution function and "CP" is the centralization proxy. This

equation provides the change in probability of protest severity equaling 2 or 3 (versus 1) when there is a

change from low to high centralization. The marginal effects of three panels of regression 8.2, taken at the

mean of all control variables, are reported in Tables 9A-9C.36 These tables report the marginal effects at

different levels of EFF and restructuring.37

[TABLES 9A, 9B, and 9C HERE]

The results reported in Tables 9A-9C are broadly consistent with the logic laid out above. Columns

9A.1, 9B.1, and 9C.1 report the marginal effects when the restructuring variable equals 0. As expected, as

the EFF variable (a "weak" shock) becomes larger, an increase in the centralization parameter decreases the

probability of a more severe protest occurring. These effects are highly signi�cant in 9A.1 and 9C.1, where

EFF borrowing of 5 times greater than the IMF quota can entail a decrease in probability of severe protest by

80% when switching from a decentralized to centralized economy. As noted before, the intuition underlying

this result is that when only weak shocks are present, centralized authorities will be better able to suppress

revolt. Moreover, it is only when the EFF variable is high enough that citizens in decentralized economies

may have incentive to protest; that is, when both shock variables are small, there is nothing to spur protest,

and there should be no difference between centralized and decentralized economies.

On the other hand, Columns 9A.3, 9B.3, and 9C. 3 report the marginal effects when the restructuring

variable equals 2.38 The marginal effects are always highly positive (between 0.878 and 0.935) and highly

signi�cant when there are two restructurings and no EFF use. Although the signi�cance goes away at higher

levels of EFF in Columns 9A.3 and 9C.3, it remains for all levels of EFF in Column 9B.3. These marginal

36The continent and year dummies are not reported at the means. The reported results are for Central America and 1985. The broad

patterns seen in Tables 9A-9C are robust to all continents and years.

37Marginal effects are derived using the nlcom procedure in Stata.

38The only irregularity in Tables 9A-9C is that the marginal effects in Column 9A.2 are negative and signi�cant when the EFF variable

equals 4 or 5 and the restructuring variable equals 1. The negative signi�cance of this variable is likely due to the omission of the

triple interaction term centralization*EFF*restructuring, which is expected to be positive. I omitted this variable because the EFF and

restructuring were too infrequently both greater than zero in order for such a variable to impact the analysis in a meaningful way.

29

effects indeed support the theory that "strong" shocks lead to an increased probability of more severe protests

happening in centralized economies.

Zambia provides an instructive case relating centralization, EFF use, and restructuring. Between 1981-

1984, Zambia borrowed signi�cantly from the EFF (3.78 times the quota) while having the lowest possible

constraint on the executive and a "bad" score (5) on political rights. Although Zambia also had loans re-

structured in two of these years, the model suggests that these relatively weak shocks would precipitate little

protest activity - and indeed, no protests occurred in these years. Following these years, Zambia restructured

or rescheduled loans in three years - 1986, 1990, and 1992. In 1986 and 1990 the constraint on the executive

was still as small as possible, while by 1992 a regime change greatly increased the constraints as well as po-

litical freedoms. The model suggests that these stronger shocks should lead to greater protest activity when

there is more centralization (1986 and 1990) but not when the economy is more decentralized (1992). In fact,

in 1986 there was a protest of severity 2 and in 1990 there was a protest of severity 3. However, when greater

constraints on executive authority and greater political rights emerged by 1992, a debt restructuring occurred

without protest.39

One possible point of contention in this analysis is that the "constraint on the executive" and "freedom"

variables are merely proxying for democracy. Indeed, both variables are positively correlated with the Polity

IV democracy index, which ranks countries from 0 (no democracy) to 10 (highly democratic). To test this,

I run robustness checks replacing the constraint on the executive and freedom variables with three different

democracy dummies - one equaling one only if the index equals 0, the second equaling one if the index equals

0 or 1, and the third equaling one if the index equals 0, 1, or 2. The results are reported in Appendix E.

Table A.1 suggests that after controlling for the presence of democracy, the number of restructurings or

reschedulings is still a strong predictor of protest, but EFF use and IMF fund use are not. This is similar

to the above results, except that EFF use is no longer a signi�cant predictor of protest. Employing the

same methodology as above, the determinants of protest severity are then tested for using the rescheduling

variables as a proxy for the "shock" and are reported in Table A.2. The only clear pattern that holds in these

regressions is that that severe protests are less likely to occur in undemocratic economies, at least when the

number of restructurings in one year is controlled for. However, this result disappears when the number of

39Of course, there are examples that contradict the results of the model and analysis. Venezuela provides one case - a severe protest

(coded 3) occurred in 1989 after a restructuring and the use of the EFF in spite of political freedoms being relatively widespread and the

presence of signi�cant constraints on the executive.

30

restructurings over 2 years is employed as the "shock" proxy. Interestingly, in 5 of the 6 regressions, the

coef�cient on the interaction term is not signi�cant. This result emerges because these regressions analyze

the effects of severe lack of democracy (0-2 on a 10 point scale) or severe constraints on the executive (0-1 on

a 7 point scale). Although there is a correlation between democracy and the constraint and freedom variables,

there is substantial variation amongst severely undemocratic countries with respect to constraints placed on

the executive and political freedoms. The regressions reported in Table A.2 suggest that the constraint and

freedom variables are indeed picking up something that the democracy variables are not - indicating that the

constraint and freedom variables are not merely proxies for the presence of democracy.

4 Massive Protest and Change in Iran

�When one person alone enacts, executes, and judges the law, there will be problems.�

�Ayatollah Javadi Amoli, June 2009

The Iranian uprising of summer 2009 typi�es the dynamics explored in the model. Massive protests

erupted in Tehran following the presidential election on June 12, which the Iranian government claimed was

won by incumbent president Mahmoud Ahmadinejad by a surprising 63-34 percent margin over moderate

challenger Mir Hussein Moussavi. Most analysts considered these results to be fraudulent, as later inspections

found that more votes were cast than there were eligible voters in numerous provinces.

The scale of these protests was enormous, with up to half a million people taking to the streets in daily

uprisings which lasted for weeks after the election results were announced. These protests occurred despite

waves of arrests and violent efforts to repress dissent � efforts revealed to the world in gruesome form in a

widely circulated video of a young woman, Neda Agha-Soltan, dying as a result of a bullet that was likely

shot by a pro-government militant.

The most surprising aspect of the revolt was that it occurred despite condemnation from the Iranian

Supreme Leader, Ayatollah Ali Khamenei. To many Iranians, the edicts of the Supreme Leader are literally

the law and are rarely questioned in public, his powers being absolute (Keller 2009; Wright 2009). The

Supreme Leader's right to rule is based on the concept of velayat-e faqih (guardianship of religious jurists),

espoused by the leader of the 1979 Revolution leader Ayatollah Ruhollah Khomeini. Theoretically, this

ensures that legislation and bills are in accordance with Islamic principles and laws, even though the Supreme

Leaders since the Revolution have frequently evaded Islamic principles in order to propagate their own power

(Tamadonfar 2001). Under this regime, the faqih is not accountable to the public, but political obedience is

31

viewed as a religious duty (Tezcür and Azadarmaki 2008). The role of the faqih in Iranian politics is codi�ed

in the Iranian Constitution, with numerous articles establishing supreme, centralized power for religious

authorities:

The powers of government in the Islamic Republic are vested in the legislature, the judiciary, and

the executive powers, functioning under the supervision of the absolute religious Leader and the

Leadership of the Ummah (italics mine)

The Constitution provides for the establishment of leadership by a holy person possessing the

necessary quali�cations and recognized as leader by the people.

The protests were a massive change from pre-election actions, where the word of the Supreme Leader was

considered and codi�ed as immutable. In spite of condemnation by Khamenei, the protests were supported

by parts of nearly every segment of Iranian society. What made the scale of this uprising so large?

The model sheds a great deal of light on the suddenness and scope of the Iranian uprising. The �central-

ized authorities�, the Supreme Leader and the religious elite, had a great deal of power over numerous types

of sanctions, as codi�ed by the Iranian Constitution and practiced by their control of the military, command

over patronage resources, and veto power over elected representatives. These institutional rules manifested

themselves in a way that the central religious authorities dictated religious, political, economic, legal, and so-

cial sanctions for anyone disobeying their dictates � not only for the citizenry, but also for political authorities

such as Ahmadinejad.

This entailed rampant preference falsi�cation prior to the elections. Indeed, a 2003 survey by Tezcür

and Azadarmaki (2008) of Tehrani views of the government revealed signi�cant dissatisfaction: 53% of

respondents said that the political system is rarely or never responsible, 45% said that the state completely

fails, and 68% said that they always or occasionally face unfair treatment by the state.

The model predicts that the centralization of authority would allow religious authorities to easily suppress

dissent following a small economic shock which affects a small portion of the population (Case 2). This was

indeed the case over the previous decade, as the central government ruthlessly suppressed student revolts in

1999 and 2003. The government tolerated these uprisings for only a few days before violently suppressing

dissent � Basij vigilantes were sent to campuses and �ung students from windows, bloodied many with bricks

or chains, and jailed many (MacFarquhar 2009).

Yet, the post-election uprisings were much more widespread, cutting across all age groups and economic

classes. This was in large part due to the nature of a national election, which is more important to the broader

32

population than the somewhat narrow concerns of the students. This sentiment was noted by one of the many

protesters, who suggested that �the student protests failed because people left the students alone. This time

they cannot defeat us because everybody has joined.� (Fathi 2009b). Even four weeks after the election results

were announced protesters lined the streets, some noting that �we are in this together� (Slackman 2009).

The widespread nature of the uprising thus conforms with the model's prediction that when shocks are

large and/or systemic, massive changes in equilibrium actions will result in centralized regimes. The model's

intuition also helps explain why the Ahmadinejad government was so concerned with preventing media cov-

erage of the protests. In the weeks after the election, the government banned coverage of the demonstrations,

arrested journalists, threatened bloggers, cut off text messaging services, and attempted to shut down social

networking websites (Fathi 2009b).40 Less media coverage meant that less Iranians were aware of the extent

of the newly-expressed, anti-authoritarian actions, hence lowering the portion of the citizens affected by the

"shock" (�).

Even before the uprising, the OpenNet Initiative (opennet.net) noted that speech was heavily regulated in

Iran and its technical �ltering system was among the most extensive in the world. In fact, the Constitution

requires that �[the media] must strictly refrain from diffusion and propagation of destructive and anti-Islamic

practices.� This makes sense in the context of the model, as the centralized nature of the government meant

that it was in its interest to suppress information that may encourage people with alternative desires to express

their opinion.

Moreover, Khamenei portrayed the unrest as the work of outsiders, especially the U.S. and Britain, claim-

ing that agents were attempting to stage a �velvet revolution� (Fathi 2009a).41 This also makes sense within

the context of the model, as the government has an interest in claiming that the protests were not truly repre-

sentative of the private preferences of Iranians. If the government were to succeed in this portrayal, it would

push social norms back to where they were before the election.

The suddenness and scale of the Iranian protests are thus attributable to two features: the centralization

of Iranian authority and the widespread affect of the seemingly fraudulent election results. The centralization

of authority entailed a situation in which preferences were falsi�ed to provide the appearance of conformity

40The regime was not completely successful in their attempt to suppress coverage of the uprising; word of the protests was broadcast to

the outside world largely through Twitter and Facebook.

41Blaming outsiders is a common tactic � for example, the Shah of Iran attributed the 1979 Revolution to a Communist conspiracy (Time

1978).

33

with the central authority, even though private opinion was vastly different. Before the election, very few

individuals were willing to transgress the authority's dictates, as doing so entailed manifold sanctions. How-

ever, the �shock� of the election results provided incentive for enough individuals to incur the costs associated

with acting in accordance with their private preferences. In turn, a cascade emerged, whereby the social pres-

sures associated with transgressing the religious authorities were loosened, and the social norm endogenously

shifted to one in which protest was acceptable. As one analyst, speaking on condition of anonymity in an

Los Angeles Times article, noted: �Public respect for [Khamenei] has been signi�cantly damaged. Opposing

him is no longer the same as opposing God� (Daragahi 2009). Unlike the student protests, whose grievances

affected a narrow swath of the population, the fraudulent nature of the election results affected all parts of the

population. In terms of the model, the signi�cance of the shock (�) and its ubiquitous nature (�) were both

large enough to encourage massive social change.

New York Times columnist Roger Cohen echoed this analysis, suggesting that �the loss of trust by millions

of Iranians who'd been prepared to tolerate a system they disliked, provided they had a small margin of

freedom, constitutes the core political earthquake in Iran� (Cohen 2009a). This is also likely the reason

why the authoritarian, centralized governments of China, Burma, and Cuba selectively censored news of the

Iranian uprising (Cha 2009). These centralized governments are similarly susceptible to cascades which arise

from �shocks�, and the Iranian uprising may have served as such a shock.

This analysis helps underscore an important feature underlying political legitimacy: legitimacy exists not

only when people believe in the legitimacy of the government, but when they believe that other people will

act in a manner that supports the legitimacy of the government. Indeed, the Iranian movement of summer

2009 suggests that legitimacy can unravel when people's beliefs about what others believe changes. There is

a subtle but important difference between this notion of legitimacy and a more conventional, Weberian notion

espoused by Razi (1987, 1990), who views legitimacy as being dependent on "the extent to which the relevant

portion of the population perceives that the regime is behaving [legitimately]" according to an established set

of norms. If this were true, then a revolt should have happened as early as the 1980s, when Khomeini made

statements undermining the norms which legitimized the theocracy (he contended that laws that violate the

Islamic law are binding only if they serve the interest of the governing elite), or when Khamenei, who was

not a leading religious scholar, was chosen to succeed Khomeini, implicitly lifting the political order over the

legitimizing religious order (Roy 1999; Tamadonfar 2001). Instead, I suggest (in line with Kuran [1995]) that

when a government acts "illegitimately" (in terms of the model, it increases i2), it only loses legitimacy when

34

citizens see others acting in a way that undermines legitimacy. Otherwise, citizens may continue to falsify

their preferences, and the government will still be viewed as legitimate.

There is signi�cant evidence that the legitimacy accorded to the Supreme Leader was undermined by the

uprising. Top religious leaders, largely encouraged by former president Ayatollah Ali Akbar Hashemi Raf-

sanjani, considered restructuring the Supreme Leader structure to a more decentralized one, openly accused

Khamenei of partisanship, met to discuss replacing Khamenei, and called the disputed presidential election

and the new government illegitimate. The most senior cleric in Iran, Grand Ayatollah Hossein Ali Montazeri,

went as far as to issue a fatwa calling Khamenei illegitimate (Sahimi 2009). Even after Ahmadinejad was

sworn in as president, a group of Iranian clerics issued an anonymous letter calling Khamenei a dictator and

demanding his removal (Worth and Fathi 2009). It is dif�cult to imagine a scenario in which such a �stunning

breaking of taboos� (MacLeod 2009) would have occurred before the uprising, as the absolute, legitimate

power wielded by Khamenei would have made such actions extremely costly.

This analysis also suggests a reason why truly democratic elections may be incompatible with centralized

authority. Two of the Iranian system's key sources of legitimacy were the principle of clerical guidance and

the popular mandate derived from democratic elections (MacLeod 2009). As one conservative Iranian cleric

noted, �the legitimacy of the election comes from the supreme leader's approval, but the level of acceptance

comes from votes� (Cohen 2009b). However, when citizens privately hold preferences that contrast with the

political leadership, private elections are one way for individuals to manifest these preferences, as doing so

does not entail any of the sanctions that would normally be incurred by a public transgression. Moreover, most

citizens have an interest, however slight, in the outcome of nationwide elections. Hence, elections provide a

rare chance for the ubiquitous breaking of norms without consequence � a situation that may result in massive

changes to equilibrium outcomes. This is why centralized regimes that attempt a façade of democracy may

abandon secret ballots or harass voters, as Robert Mugabe did in the 2008 Zimbabwe presidential election, or

they may fabricate the results to such an extent that the belief of a democratic system never really emerges, as

was the case in Saddam Hussein's Iraq (where Hussein won two different elections with 99.96% and 100%

of the vote). The former strategy maintains the costs imposed by transgressing (not voting for) the authority,

thus decreasing the likelihood of a shock, while the latter strategy entails that the election �results� are not a

shock.

Although the Iranian protests did not lead to a regime change - Ahmadinejad ended up being sworn in

by Khamenei for a second term - they did fundamentally alter Iranian political and religious institutions and

35

the legitimizing relationship between the two. The Ahmadinejad wing was able to maintain power because it

controlled the military, not because it was legitimized by religious authorities. This encouraged Ahmadinejad

and his Basij supporters to enact a more militant, authoritarian regime � legitimized by force, not religion.

We can only speculate what this means for the political future of Iran, but history � and the model � suggest

that an even larger movement may not be too far off.

5 Conclusion

This paper analyzes the institutional roots of massive social, economic, and political changes. It employs

a simple economic model which suggests that authorities in centralized economies are relatively insulated

from small economic shocks but may be susceptible to massive changes if economic shocks are larger. This

result arises because citizens in centralized economies are more likely to falsify their preferences, as they

face numerous costs from choosing actions near their intrinsic optima, and centralized authorities have less

incentive to accommodate citizens than decentralized authorities. This entails that small economic shocks

are unlikely to upset the equilibrium outcome, as individuals are unwilling to incur the numerous sanctions

associated with transgressing the centralized authority's dictates in response to a small shock. However, a

large shock may encourage some citizens to incur these costs, which can result in a cascade whereby social

norms, and eventually institutional dictates, dramatically change.

This hypothesis is tested using data from austerity protests in 1976-1992. I use various measures of

IMF involvement to proxy for a "shock", as such involvement often directly led to austerity measures that

contributed to existence and severity of the protests. A regression analysis supports the model's hypothesis,

suggesting that after an increase in a "weaker" IMF variable (use of the Extended Fund Facility), protests are

more severe in decentralized economies. Yet, an increase in a "stronger" IMF variable (the number of debt

restructurings or renegotiations) led to more severe protests in more centralized economies.

This logic is also employed to shed light on aspects of the Iranian protests of summer 2009. It suggests

that the scale of the uprising can be attributed to the centralization of coercive power and the widespread affect

of the election results. The latter entailed that the "shock" was widespread throughout the population, whereas

the former entailed that falsi�cation of beliefs about the government was rampant before the elections. These

two phenomena combined to form a cascade of opinion after the election results were announced, resulting

in a massive change in publicly expressed opinion.

The model and the insights drawn from the Iranian case also helps shed new light on the underpinnings

36

of political legitimacy. Namely, it suggests that legitimacy is an equilibrium outcome only when individuals

believe that other people will act in a manner that supports the legitimacy of the government. Legitimacy can

thus be fragile in centralized regimes, as high levels of private dissatisfaction entail that a shock can lead to a

signi�cant change in people's beliefs about what others believe.

The model applies to many other historical and contemporary phenomena. For example, the fall of

communism in the Soviet Union, due in part to the multi-faceted sanctions imposed by the Communist Party,

may in part be explained by the logic spelled out in this model. Similarly, the centralization of authority

in Tsarist Russia may have led to the incredible scale of the Bolshevik Revolution (while also explaining

the central government's ability to suppress numerous smaller outbreaks in the nineteenth century), and the

interdependence between the Chinese bureaucracy and Qing rulers may help explain the scale of the Taiping

Rebellion (1850-1864), which followed after massive de�ation systemically affected much of the Chinese

countryside. Moreover, the model may have implications for the future of centralized regimes such as those

in Iran, China, Burma (Myanmar), North Korea, and Zimbabwe.42

This paper is not meant to suggest that a general framework exists for predicting revolutions or massive

social change.43 The amount of variables necessary for a revolution to occur likely renders this an impossible

task. This paper, does, however, suggest that one ubiquitous set of conditions - those associated with the

centralization of coercive power - are conducive to massive changes in equilibrium outcomes. In a sense, then,

the conclusion is much different fromKuran (1995), who argues that revolutions are completely unpredictable

based on structural conditions. Instead, this paper argues - using a framework similar to Kuran's - that certain

institutional conditions are more suitable for a revolution or massive equilibrium change than others.

References

Akerlof, George A. 1980. "A Theory of Social Custom, ofWhich Unemployment May be One Consequence."

Quarterly Journal of Economics 94(4): 749-775.

42Of course, the model also has implications for decentralized regimes, especially ones espousing separation between church and state.

Such regimes are less likely to be subject to revolutions, as decentralized authorities have more incentive to respond to the citizenry, thus

reducing the total level of preference falsi�cation.

43Nor is this paper meant to lay out any speci�c prediction concerning the future of centralized economies, à la Bruce Bueno de Mesquita.

Instead, it provides a general framework which links centralized institutions to revolutionary activity. For a particularly relevant example

of political forecasting, see Bueno de Mesquita (1984).

37

����. 1997. "Social Distance and Social Decisions." Econometrica 65(5): 1005-1027.

Akerlof, George A., and Rachel E. Kranton. 2000. "Economics and Identity."Quarterly Journal of Economics

115(3): 715-753.

����. 2005. "Identity and the Economics of Organizations." Journal of Economic Perspectives 19(1):

9-32.

Anderson, Lisa R., and Charles A. Holt. 1997. "Information Cascades in the Laboratory." American Economic

Review 87(5): 847-862.

Auvinen, Juha. 1996. �IMF Intervention and Political Protest in the Third World: A Conventional Wisdom

Re�ned.� Third World Quarterly 17(3): 377-400.

__________. 1997. �Political Con�ict in Less Developed Countries, 1981-89.� Journal of Peach Research

34(2): 177-195.

Banerjee, Abhijit V. 1992. "A Simple Model of Herd Behavior." Quarterly Journal of Economics 107(3):

797-817.

Barrett, D.B., G.T. Kurian, and T.M. Johnson 2001). World Christian Encyclopedia, 2nd ed. Oxford: Oxford

University Press.

Barro, Robert J. and Rachel M. McCleary. 2005. �Which Countries Have State Religions?�Quarterly Journal

of Economics 120(4): 1331-1370.

Bernheim, B. Douglas. 1994. "A Theory of Conformity." Journal of Political Economy 102(5): 841-877.

Bikhchandani, Sushil, David Hirshleifer, and Ivo Welch. 1992. "A Theory of Fads, Fashion, Custom, and

Cultural Change as Information Cascades." Journal of Political Economy 100(5): 992-1026.

����. 1998. "Learning from the Behavior of Others: Conformity, Fads, and Informational Cascades."

Journal of Economic Perspectives 12(3): 151-170.

Bueno de Mesquita, Bruce. 1984. "Forecasting Policy Decisions: An Expected Utility Approach to Post-

Khomeini Iran." PS 17(2): 226-236.

Cha, Ariana Eunjung. �Authoritarian Regimes Censor News in Iran.�Washington Post, June 27, 2009.

38

Cohen, Roger. �Iran's Children of Tomorrow.� New York Times, June 23, 2009a.

����. �Let the Usurpers Writhe.� New York Times, July 1, 2009b.

Cosgel, Metin M. and Thomas J. Miceli. 2009. "State and Religion." Journal of Comparative Economics

37(3): 402-416.

Cosgel, Metin M., Thomas J. Miceli, and Rasha Ahmed. 2009. "Law, state power, and taxation in Islamic

history." Journal of Economic Behavior & Organization 71(3): 704-717.

Cosgel, Metin M., Thomas J. Miceli, and Jared Rubin. 2010a. "Guns and Books: Legitimacy and Technolog-

ical Change in the Ottoman Empire and Europe." Working Paper.

����. 2010b. "Political Legitimacy and Technology Adoption." Working Paper.

Çelen, Bo�gaçhan, and Shachar Kariv. 2004. "Distinguishing Informational Cascades from Herd Behavior in

the Laboratory." American Economic Review 94(3): 484-498.

����. 1993. "The Economics of Rumours." Review of Economic Studies 60(2): 309-327.

Daragahi, Borzou. �Supreme leader Khamenei diminished in Iranians' eyes.� Los Angeles Times, July 15,

2009.

Davies, James C. 1962. "Toward a Theory of Revolution." American Sociological Review 27(1): 5-19.

Dillon, K. Burke, C. Maxwell Watson, G. Russell Kincaid, and Chanpen Puckahtikom. 1985. Recent Devel-

opments in External Debt Restructuring. Washington: IMF.

Fathi, Nazila. �Iran's Top Leader Dashes Hopes for a Compromise.� New York Times, June 20, 2009a.

����. �Protesters Defy Iranian Efforts to Cloak Unrest.� New York Times, June 18, 2009b.

Fershtman, Chaim, Kevin M. Murphy, and Yoram Weiss. 1996. "Social Status, Education, and Growth."

Journal of Political Economy 104(1): 108-132.

Fershtman, Chaim, and Yoram Weiss. 1993. "Social Status, Culture and Economic Performance." Economic

Journal 103: 946-959.

Freedom House. 2009. Freedom in the World. Washington: Freedom House.

Gintis, Herbert. 2003. "Solving the Puzzle of Prosociality." Rationality and Society 15(2): 155-187.

39

Goldstone, Jack A. 1994. "Is Revolution Individually Rational?" Rationality and Society 6(1): 139-166.

����. 2001. "Toward a Fourth Generation of Revolutionary Theory." Annual Review of Political Science

4: 139-187.

Greif, Avner. 2002. "The Islamic Equilibrium: Legitimacy and Political, Social, and Economic Outcomes,"

Working Paper, Stanford University.

����. 2005. "Commitment, Coercion, and Markets: The Nature and Dynamics of Institutions Supporting

Exchange", in (C.Menard andM.M. Shirley, eds.),Handbook for New Institutional Economics, Norwell,

MA: Kluwer Academic Publishers.

����. 2006. Institutions and the Path to the Modern Economy: Lessons from Medieval Trade. Cambridge:

Cambridge University Press.

����. 2009. "Morality and Institutions." Mimeo, Stanford University.

����. 2010. "A Theory of Moral Persistence: The Role of Crypto-morality." Journal of Comparative

Economics: forthcoming.

Iannaccone, Laurence R. 1988. "A Formal Model of Church and Sect." American Journal of Sociology

94(Supplement): S241-S268.

International Monetary Fund. (Various). Annual Report. Washington: IMF.

Iyigun, Murat. 2009. "Monotheism (From a Sociopolitical & Economic Perspecitve)." Working Paper.

Karaman, Kamil Kivanç. 2009. "Decentralized coercion and self-restraint in provincial taxation: The Ot-

toman Empire, 15th�16th centuries." Journal of Economic Behavior & Organization 71(3): 690-703.

Keller, Bill. �Innocent Googling? No Such Thing in Tehran.� New York Times, June 17, 2009.

Klein, Naomi. 2007. Shock Doctrine: The Rise of Disaster Capitalism. New York: Holt and Company.

Kuhn, Michael G., and Jorge P. Guzmàn. 1990. Multilateral Of�cial Debt Rescheduling: Recent Experience.

Washington: IMF.

Kuran, Timur. 1987. "Preference Falsi�cation, Policy Continuity and Collective Conservatism." Economic

Journal 97: 642-665.

40

����. 1989. "Sparks and prairie �res: A theory of unanticipated political revolution." Public Choice 61:

41-71.

����. 1995a. Private Truths, Public Lies: The Social Consequences of Preference Falsi�cation. Cam-

bridge: Harvard University Press.

����. 1995b. "The Inevitability of Future Revolutionary Surprises." American Journal of Sociology

100(6): 1528-1551.

����. 1998. "Ethnic Norms and Their Transformation through Reputational Cascades." Journal of Legal

Studies 27(2): 623-659.

����. 2005. "The Absence of the Corporation in Islamic Law: Origins and Persistence." The American

Journal of Comparative Law 53: 785-834.

Kuran, Timur, and William H. Sandholm. 2008. "Cultural Integration and Its Discontents." Review of Eco-

nomic Studies 75: 201-228.

Kuran, Timur, and Cass R. Sunstein. 1999. "Availability Cascades and Risk Regulations." Stanford Law

Review 51(4): 683-768.

Lewis, Bernard. 1974. Islam: From the Prophet Muhammad to the Capture of Constantinople. New York:

Harper & Row.

����. 1993. Islam and the West. Oxford: Oxford University Press.

����. 1995. The Middle East. New York: Scribner.

MacFarquhar, Neil. �Ïran's Latest Protests Are Seen as the Toughest to Stop.� New York Times, June 17, 2009.

MacLeod, Scott. �Iran's Opposition Down but Not Out.� Time.com, July 1, 2009.

Marshall, Monty G. and Keith Jaggers. 2008. Polity IV project: Political regime characteristics and transi-

tions, 1800-2008.

Norton, Edward C., Hua Wang, and Chunrong Ai. 2004. �Computing interaction effects and standard errors

in logit and probit models.� The Stata Journal 4(2): 154-167.

Olson, Mancur. 1963. "Rapid Growth as a Destabilizing Force." Journal of Economic History 23(4): 529-552.

41

����. 1971. The Logic of Collective Action: Public Goods and the Theory of Groups. Cambridge: Harvard

University Press.

Roncek, Dennis W. 1992. �Learning More From Tobit Coef�cients: Extending a Comparative Analysis of

Political Protests.� American Sociological Review 57(4): 503-507.

Razi, G. Hossein. 1987. "The Nexus of Legitimacy and Performance: The Lessons of the Iranian Revolution."

Comparative Politics 19(4): 453-469.

����. 1990. "Legitimacy, Religion, and Nationalism in the Middle East." American Political Science

Review 84(1): 69-91.

Romer, David. 1984. "The Theory of Social Custom: A Modi�cation and Some Extensions." Quarterly

Journal of Economics 99(4): 717-727.

Roy, Olivier. 1999. "The Crisis of Religious Legitimacy in Iran."Middle East Journal 53(2): 201-216.

Rubin, Jared. 2009. "Institutions, the Rise of Commerce, and the Persistence of Laws: Interest Restrictions

in Islam & Christianity." Working paper, California State University, Fullerton.

Sahimi, Muhammad. �Grand Ayatollah Montazeri's Fatwa.� Tehran Bureau, July 12, 2009.

Shugart, Matthew Soberg. 1989. "Patterns of revolution." Theory and Society 18: 249-271.

Slackman, Michael. �Iran Protesters Take to Streets Despite Threats.� New York Times, July 9, 2009.

Tamadonfar, Mehran. "Islam, Law, and Political Control in Contemporary Iran." Journal for the Scienti�c

Study of Religion 40(2): 205-219.

Tanter, Raymond, andManusMidlarsky. 1967. "A theory of revolution." Journal of Con�ict Resolution 11(3):

264-280.

Tezcür, Günes Murat, and Taghi Azadarmaki. 2008. "Religiosity and Islamic Rule in Iran." Journal for the

Scienti�c Study of Religion 47(2): 211-224.

Time Magazine. �The Shah's Divided Land.� Time, September 18, 1978.

Tullock, Gordon. 1971. "The Paradox of Revolution." Public Choice 1: 89-99.

42

Walton, John and Charles Ragin 1990. �Global and National Sources of Political Protest: Third World Re-

sponses to Debt Crisis.� American Sociological Review 55(6): 876-890.

Walton, John and David Seddon. 1994. Free Markets & Food Riots. Cambridge, MA: Blackwell.

World Bank. (Various).World Development Indicators. Washington: World Bank.

����. (Various).World Tables. Baltimore: Johns Hopkins University Press.

Worth, Robert F., and Nazila Fathi. "Clerics' Call for Removal Challenges Iran Leader." New York Times,

August 16, 2009.

Wright, Robin. �Iran's Embattled Supreme Leader: A Test for Khamenei.� Time.com, June 25, 2009.

43

A Proofs

A.1 Proof of Proposition 1

In order to determine the effect of dependence on the level of social change (�), �rst determine the sign of

the derivative @�@ . Note that:@�

@ =@a0@

� @at�

@ :

Denoting � = b+ c f+ gjj+k

f+ g+h +djj+k , start by solving for

@a0@ :

@a0@

=

cg�

0@ jiB2j+k (f+ g+h)�

�fiB1 +

gjiB2j+k

�(f+ g+h)2

1A� cg� jj+k (f+ g+h)�(f+

gjj+k )

(f+ g+h)2

� cfiB1 +

gjiB2j+k

f+ g+h +djiB2j+k +

bN

Pj a

Bj

!�2

=

�jf�iB2 � iB1

�+ jhiB2 � kfiB1

��� (jh� kf)

cfiB1 +

gjiB2j+k

f+ g+h +djiB2j+k +

bN

Pj a

Bj

!(j+k)(f+ g+h)2

cg �2

=jf�iB2 � iB1

��+ jhcf

f+ g+h

�iB2 � iB1

�+�jkfj+k

��c g

f+ g+h + d� �iB2 � iB1

�(j+k)(f+ g+h)2

cg �2

+bhjh�iB2 � 1

N

Pj a

Bj

�� kf

�iB1 � 1

N

Pj a

Bj

�i(j+k)(f+ g+h)2

cg �2

=jf�iB2 � iB1

�[b+ c+ d] + b

hjh�iB2 � 1

N

Pj a

Bj

�� kf

�iB1 � 1

N

Pj a

Bj

�i(j+k)(f+ g+h)2

cg

�b+

cf+ c gjj+k

f+ g+h +djj+k

�2 (A3)

Next, solve for @at@ , denoting = b+c+d+e, � = b+cf+ c gj

j+k

f+ g+h +djj+k , and = +(1� �) b (� � 1):

@at@

=

0BBBBBBBB@

cg [�b (� � 1) + ( + (1� �) b (� � 1))�]

0@ (f+ g+h)

�jiB2j+k

���fiB1 +

gjiB2j+k

�(f+ g+h)2

1A�cg

�(f+ g+h)( j

j+k )�(f+ gjj+k )

(f+ g+h)2

�2664

cfiB1 +

c gjiB2j+k

f+ g+h +djiB2j+k

!+

�bN

Pj;S a

Bj +

(+b(��1))bN

Pj;NS a

Bj

3775

1CCCCCCCCA[�b (� � 1) + ( + (1� �) b (� � 1))�]2

44

=

0BB@�jf�iB2 � iB1

�+ jhiB2 � kfiB1

�[�b (� � 1) + �]�

(jh� kf)"

cfiB1 +

c gjiB2j+k

f+ g+h +djiB2j+k

!+ �b

N

Pj;S a

Bj +

(+b(��1))bN

Pj;NS a

Bj

#1CCA

(j+k)(f+ g+h)2

cg [�b (� � 1) + ( + (1� �) b (� � 1))�]2

=

0B@ �jf�iB2 � iB1

�(b+ c+ d) + b

�jhiB2 � kfiB1

��+

�b (� � 1)�jf�iB2 � iB1

�+ jhiB2 � kfiB1

�� b (jh� kf)

��N

Pj;S a

Bj +

+b(��1)N

Pj;NS a

Bj

�1CA

(j+k)(f+ g+h)2

cg [�b (� � 1) + ( + (1� �) b (� � 1))�]2

=

0BBBB@jf�iB2 � iB1

�[�b (� � 1) + ( + (1� �) b (� � 1)) (b+ c+ d)]

b�hjh��iB2 � 1

N

Pj;S a

Bj

�� kf

��iB1 � 1

N

Pj;S a

Bj

�i+

b ( + b (� � 1))hjh�(1� �) iB2 � 1

N

Pj;NS a

Bj

�� kf

�(1� �) iB1 � 1

N

Pj;NS a

Bj

�i1CCCCA

(j+k)(f+ g+h)2

cg(+(1��)b(��1))

��b (� � 1) + ( + (1� �) b (� � 1))

�b+ c

f+ gjj+k

f+ g+h +djj+k

��2Combining these two, this means that @E@ > 0 when (again denoting denoting = b + c + d + e,

� = b+cf+ c gj

j+k

f+ g+h +djj+k , and = + (1� �) b (� � 1)):

jf�iB2 � iB1

�(� e) + b

hjh�iB2 � 1

N

Pj a

Bj

�� kf

�iB1 � 1

N

Pj a

Bj

�i(j+k)(f+ g+h)2

cg �2

>

0BBBB@jf�iB2 � iB1

�[�b (� � 1) + (� e)]

b�hjh��iB2 � 1

N

Pj;S a

Bj

�� kf

��iB1 � 1

N

Pj;S a

Bj

�i+

b ( + b (� � 1))hjh�(1� �) iB2 � 1

N

Pj;NS a

Bj

�� kf

�(1� �) iB1 � 1

N

Pj;NS a

Bj

�i1CCCCA

(j+k)(f+ g+h)2

cg [�b (� � 1) + �]2

=)

0B@ �22b2 (� � 1)2+

2�b (� � 1)� + 2�2

1CA0B@ jf

�iB2 � iB1

�(� e)+

bhjh�iB2 � 1

N

Pj a

Bj

�� kf

�iB1 � 1

N

Pj a

Bj

�i1CA

> �2

0BBBBBBBBBBBB@

jf�iB2 � iB1

�[�b (� � 1) + (� e)]+

b�

264 jh��iB2 � 1

N

Pj;S a

Bj

��

kf��iB1 � 1

N

Pj;S a

Bj

�375+

b ( + b (� � 1))

264 jh�(1� �) iB2 � 1

N

Pj;NS a

Bj

��

kf�(1� �) iB1 � 1

N

Pj;NS a

Bj

�375

1CCCCCCCCCCCCA

45

=) �b (� � 1) (�b (� � 1) + 2�)

0B@jf �iB2 � iB1 � (� e) + b264 jh

�iB2 � 1

N

Pj a

Bj

��

kf�iB1 � 1

N

Pj a

Bj

�3751CA

> b�2

0BBBBBBBBBBBB@

jf�iB2 � iB1

��(� � 1) + �

264 jh��iB2 � 1

N

Pj;S a

Bj

��

kf��iB1 � 1

N

Pj;S a

Bj

�375+

( + b (� � 1))

264 jh�(1� �) iB2 � 1

N

Pj;NS a

Bj

��

kf�(1� �) iB1 � 1

N

Pj;NS a

Bj

�375�

�jh�iB2 � 1

N

Pj a

Bj

�� kf

�iB1 � 1

N

Pj a

Bj

��

1CCCCCCCCCCCCA

=) �(�b (� � 1) + 2�)

0B@jf �iB2 � iB1 � (� e) + b264 jh

�iB2 � 1

N

Pj a

Bj

��

kf�iB1 � 1

N

Pj a

Bj

�3751CA

> �2

0BBBBBBBB@jf�iB2 � iB1

��+ (� b (1� �))

264 jh��iB2 � 1

N

Pj;S a

Bj

��

kf��iB1 � 1

N

Pj;S a

Bj

�375+

�b

264 jh�(1� �) iB2 � 1

N

Pj;NS a

Bj

��

kf�(1� �) iB1 � 1

N

Pj;NS a

Bj

�375

1CCCCCCCCA

=)��2b (� � 1) + 2�� �2

�0BBBB@

�jf�iB2 � iB1

�(� e)+

b

264 jh��iB2 � �

N

Pj a

Bj

��

kf��iB1 � �

N

Pj a

Bj

�3751CCCCA (A4)

> �2

0@�jf �iB2 � iB1 � e+ (� b)24jh

0@�iB2 � 1

N

Xj;S

aBj

1A� kf0@�iB1 � 1

N

Xj;S

aBj

1A351AHence, if (A4) holds, @�@ > 0. The rest of the proof shows that under some conditions (A4) is always

true, and under other conditions, (A4) is only true when � is suf�ciently large. (A4) is only true when � is

suf�ciently large when 1) (A4) does not hold at � ! 1; 2) (A4) holds at � ! 1; 3) the derivative of the

left-hand side of (A4) with respect to � is greater than the derivative of the right hand side with respect to �.

On the other hand, if 2 and 3 hold but (A4) is also true at � ! 1, then (A4) always holds. Start by analyzing

46

(A4) at � ! 1, where = b+ c+ d+ e, � = b+ cf+ c gjj+k

f+ g+h +djj+k , and = + (1� �) b (� � 1):

0BBBBBBBBBBBB@

jf�iB2 � iB1

�(2 (� e)� �)+

b (2� �)

264 jh�iB2 � 1

N

Pj a

Bj

��

kf�iB1 � 1

N

Pj a

Bj

�375�

� (� b)

264 jh�iB2 � 1

�N

Pj;S a

Bj

��

kf�iB1 � 1

�N

Pj;S a

Bj

�375

1CCCCCCCCCCCCA> 0 (A5)

Next, (A4) holds at � !1 if:

��2 + 2 (1� �)�� (1� �) �2

�0BBBB@

�jf�iB2 � iB1

�(� e)+

b

264 jh��iB2 � �

N

Pj a

Bj

��

kf��iB1 � �

N

Pj a

Bj

�3751CCCCA (A6)

> (1� �) �2

0B@�jf �iB2 � iB1 � e+ (� b)264 jh

��iB2 � 1

N

Pj;S a

Bj

��

kf��iB1 � 1

N

Pj;S a

Bj

�3751CA

=) �jf�iB2 � iB1

�[(1� �) � (2 (� e)� �) + �(� e)] >

(1� �) �2 (� b)

264 jh��iB2 � 1

N

Pj;S a

Bj

��

kf��iB1 � 1

N

Pj;S a

Bj

�375�

b��2 + (1� �) � (2� �)

�264 jh��iB2 � �

N

Pj a

Bj

��

kf��iB1 � �

N

Pj a

Bj

�375

=)

0BBBBBBBBBBBB@

jf�iB2 � iB1

�(2 (� e)� �)+

b (2� �)

264 jh�iB2 � 1

N

Pj a

Bj

��

kf�iB1 � 1

N

Pj a

Bj

�375�

� (� b)

264 jh�iB2 � 1

�N

Pj;S a

Bj

��

kf�iB1 � 1

�N

Pj;S a

Bj

�375

1CCCCCCCCCCCCA>

��2(1� �) �

0BBBB@jf�iB2 � iB1

�(� e)+

b

264 jh�iB2 � 1

N

Pj a

Bj

��kf

�iB1 � 1

N

Pj a

Bj

�3751CCCCA

(A7)

Note that the left-hand side of (A7) is equal to the left-hand side of (A5). Finally, check whether the derivative

of the left-hand side of (A4) with respect to � is greater than the derivative of the right hand side with respect

47

to �. This is true when:

��2 + 2 (1� �)�� (1� �) �2

�0BBBB@

�jf�iB2 � iB1

�(� e)+

b

264 jh��iB2 � �

N

Pj a

Bj

��kf

��iB1 � �

N

Pj a

Bj

�3751CCCCA (A8)

> (1� �) �2

0B@�jf �iB2 � iB1 � e+ (� b)264 jh

��iB2 � 1

N

Pj;S a

Bj

��kf

��iB1 � 1

N

Pj;S a

Bj

�3751CA

This condition is the same one as in (A6). Hence, @�@ is always greater than 0 if (A5) and (A7) hold, but is

greater than 0 only at suf�ciently large � if (A7) holds but (A5) does not. Denote � as the left-hand side of

(A7) and � as the right-hand side of (A7) : Hence, we have the following three cases:

Case 1: @�@ is always greater than 0. This holds if � > 0.

Case 2: @�@ is always less than 0. This holds if � < �.

Case 3: There exists some �� such that @�@ < 0 when � < �� and @�

@ > 0 when � > ��. This holds if

� 2 (�; 0).

Next, determine where these cases fall in the �� iB2 plane. The left-hand side of (A5), �, is constant in

�. To see how � changes in iB2 , take the derivative:

@�

@iB2= jf (2 (� e)� �)�jh (�� 2b)

This must be positive since jf > jh and 2 (� e)� � > �� 2b. Thus, an increase in iB2 increases the

parameter set over which Case 1 emerges. Case 1 will thus always emerge if � > 0 at the minimum value of

iB2 . By assumption, the lowest value of iB2 = iB1 . At iB2 = iB1 , � equals:

� = (jh� kf)

0B@ b (2� �)�iB2 � 1

N

Pj a

Bj

��

� (� b)�iB2 � 1

�N

Pj;S a

Bj

�1CA

This is de�nitely negative if� (� b) > b (2� �)() � > 2b. For the remainder of the analysis, assume

that this condition holds. In words, this simply means that citizens care suf�ciently about sanctions imposed

by the authorities relative to their own utility, as discussed in the body of the text. With this assumption,

� < 0 at iB2 = iB1 , and thus there must exist some i� where � � 0 and Case 1 emerges when iB2 � i� and

48

� < 0 when < i�. i� is de�ned as follows:

i� =

iB1

264 jf (2 (� e)� �)�

kf (�� 2b)

375+ (jh� kf)264 b (2� �)

�1N

Pj a

Bj

��

� (� b)�

1�N

Pj;S a

Bj

�375

jf (2 (� e)� �)�jh (�� 2b) (A9)

When iB2 < i�, either Case 2 or Case 3 is possible, depending on the value of �. Case 2 emerges when

� < �. Since @�@iB2

> 0 and @�@iB2

< 0, it must be that the parameter set over which Case 2 emerges is weakly

decreasing in iB2 . It is also clear that � ! 0� as � ! 0 and � ! �1 as � ! 1. This means that when

iB2 < i�, Case 2 emerges when � ! 0 and Case 3 emerges when � ! 1. Moreover, since @�@� < 0, this

means that there must be some �� for every value of iB2 (< i�) where � < �� entails that Case 2 emerges

and � � �� entails that Case 3 emerges. Finally, the line dividing the space between Cases 2 and 3 must be

concave since @2�@�2 > 0.

Finally, if Case 3 is the relevant one, it is clear that any change in a parameter which increases the

likelihood of Case 1 emerging decreases ��, while a change in a parameter which increases the likelihood of

Case 2 emerging increases ��. Thus, �� is decreasing in � and iB2 .

49

B Tables

1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 TotalAlgeria X X 2Argentina X X X X X X X X X X X X 12Bangladesh X X X X X X X X X X X X X X X X X 17Benin X X X X X 5Bolivia X X X X X X X X X X X X X X 14Brazil X X X X X X X X X X X 11Bulgaria X X X 3Burkina Faso X X X 3Cameroon X X X X X X X X X X X X X X X X 16Central African Republic X X X X X X X X X X X X 12Chad X X X 3Chile X X X X X X X X X X X X X X X X X 17Congo (Dem. Rep.) X X X X X X X X X X X X X X X X 16Congo (Rep.) X X X X X X X X X X X X X X X 15Costa Rica X X X X X X X X X X X X X 13Côte d'Ivoire X X X X X X X X X X X X X 13Dominican Republic X X X X X X X X X X X X X X X 15Ecuador X X X X X X X X X X X X 12Egypt X X X X X X X X X X X X X X X X X 17El Salvador X X X X X X X X X X X 11Ethiopia X X X X X X X X 8Gabon X X X X X X X X X X X X X X X X 16Gambia, The X X X X X X X X X X X X X X X X 16Ghana X X X X X X X X X X X 11Guatemala X X X X X 5Guinea X X X X X X X X 8Guinea­Bissau X X X X X 5Guyana X X X X X X X X X X X X X X X X 16Haiti X X X X X X X X X X X X 12Honduras X X X X X X X X X X X X X X X 15Hungary X X X 3India X X X X X X X X X X X X X 13Iran 0Jamaica X X X X X X X X X X X X X X X X X 17Jordan X X X X X 5Kenya X X X X X X X X X X X X X X X X X 17Korea (South) 0Madagascar X X X X X X X X X X X X X X 14Malawi X X X X X X X X X X X X X X X 15Mali X X X X X X X X X X X X 12Mauritania X X X X X X X X X X X X X X X X 16Mauritius X X X X X X X X X X X X 12Mexico X X X X X X X X X X X X X X X X X 17Morocco X X X X X X X X X X X X X X X X 16Mozambique X X X X X X X X X X 10Nepal X X X 3Nicaragua X X X X X X X X X X X X X X X 15Niger X X X X X X X X X X X 11Nigeria X X X X X X X X X X X 11Pakistan X X X X X X X X X X X X X X X X X 17Panama X X X X X X X X X X X X 12Papua New Guinea X X X X X 5Peru X X X X X X X X X X X X X X X X X 17Philippines X X X X X X X X X X X X X X X X X 17Poland X X X X X X X X X X X X X 13Romania X X X X X X X X X X X X X X X X X 17Senegal X X X X X X X X X X X X X X X X 16Sierra Leone X X X X X X X X X X X X X X X X X 17Sri Lanka X X X X X X X X X X X X X X X X X 17Sudan X X X X X X X X X X X X X X X X X 17Tanzania X X X X X X X X X X X X X X X X X 17Thailand X X X X X X X X X X X X X 13Togo X X X X X X X X X X X X X X X 15Trinidad and Tobago X X X X 4Tunisia X X X X X X X 7Turkey X X X X X X X X X X X X X 13Uruguay X X X X X X X X X X X X X X X X X 17Venezuela X X X X X X X X X X X X 12Zambia X X X X X X X X X X X X X X X X X 17Zimbabwe X X X X X X X X X X X 11

Table 1: IMF Pressure by Country and Year

X = IMF Pressure within one year

50

CountryProtests

Indexed 1Protests

Indexed 2Protests

Indexed 3Total

ProtestsAlgeria 0 1 1 2Argentina 0 1 0 1Bangladesh 1 0 0 1Bolivia 5 2 0 7Brazil 3 5 0 8Bulgaria 1 0 0 1Chile 1 3 0 4Congo (Dem. Rep.) 1 0 1 2Côte d'Ivoire 0 0 1 1Dominican Republic 4 2 1 7Ecuador 3 1 0 4Egypt 0 0 1 1El Salvador 2 0 0 2Gabon 1 0 0 1Ghana 1 0 0 1Guatemala 0 1 0 1Haiti 0 1 0 1Honduras 1 0 0 1India 6 0 0 6Iran 1 0 0 1Jamaica 2 2 0 4Jordan 0 0 1 1Mexico 4 0 0 4Morocco 0 1 2 3Mozambique 1 0 0 1Nicaragua 1 0 0 1Niger 0 1 0 1Nigeria 0 3 0 3Panama 2 2 0 4Peru 3 9 0 12Philippines 3 0 0 3Poland 3 0 0 3Romania 3 0 2 5Senegal 1 0 0 1Sierra Leone 0 1 0 1South Korea 0 0 1 1Sudan 2 3 0 5Tunisia 1 0 0 1Venezuela 2 2 1 5Zambia 2 1 1 4Total 61 42 13 116

Table 2: Protests And Severity by Country

51

YearProtests

Indexed 1Protests

Indexed 2Protests

Indexed 3Number of

Protests1976 0 1 0 11977 0 2 1 31978 1 2 0 31979 0 1 0 11980 0 1 1 21981 1 0 1 21982 0 3 0 31983 7 5 0 121984 7 3 2 121985 6 5 0 111986 7 5 0 121987 5 2 0 71988 2 2 1 51989 8 1 3 121990 4 6 2 121991 6 2 2 101992 7 1 0 8Total 61 42 13 116

Table 3: Protests and Severity by Year

RegionProtests

Indexed 1Protests

Indexed 2Protests

Indexed 3Total

ProtestsAfrica 9 9 3 21Asia 10 0 1 11Central America 16 8 1 25Europe 7 0 2 9Middle East 2 2 5 9South America 17 23 1 41Total 61 42 13 116

Table 4: Protests and Severity by Continent

52

Observations Mean VarianceOnly Including Years Used in Regressions

IMF and Centralization VariablesUse of Funds/Quota (if > 1.25) 216 0.58 1.076Extended Fund Facility/Quota 133 0.51 1.614# of reschedulings/restructurings 274 0.37 0.289Constraint on Executive (1­7) 807 3.40 4.761Political Rights (1­7) 807 4.45 3.608

Economic and Demographic ControlsUrbanization Rate (%) 807 39.71 389.964Real GDP (per capita, $K, 1987 USD) 803 0.96 0.844Population (millions) 807 31.58 8448.2Religious Fractionalization 807 0.71 0.047

Protest Years OnlyUse of Funds/Quota (if > 1.25) 40 1.04 1.383Extended Fund Facility/Quota 19 0.81 2.454# of reschedulings/restructurings 43 0.59 0.394Constraint on Executive (1­7) 81 4.09 5.430Political Rights (1­7) 81 3.63 3.111

Economic and Demographic ControlsUrbanization Rate 81 51.34 341.627Real GDP (per capita, $K, 1987 USD) 81 1.15 0.713Population (millions) 81 60.95 25468.7Religious Fractionalization 81 0.81 0.030

Table 5: Summary Statistics, IMF and Centralization Variables

**Years used in regressions are ones within one year of one of the IMF variables being positive

53

DEPENDENT VARIABLE: Protest Presence (0 or 1)

(6.1) (6.2) (6.3)

Constraint on ExecutiveDummy 1

Constraint on ExecutiveDummy 2 Political Rights Dummy

IMF Pressure Indices

Use of Funds/Quota (if > 1.25) (δ1) 0.245* 0.135 0.12[0.129] [0.145] [0.134]

Interaction with Centralization Variable (δ3) ­0.994*** ­0.194 ­0.234[0.319] [0.232] [0.249]

Extended Fund Facility/Quota (δ1) 0.290** 0.325*** 0.214**[0.114] [0.125] [0.108]

Interaction with Centralization Variable (δ3) ­0.911* ­0.444* ­0.63[0.541] [0.235] [0.542]

# of reschedulings/restructurings (δ1) 0.550** 0.570** 0.595**[0.259] [0.279] [0.263]

Interaction with Centralization Variable (δ3) 0.047 ­0.020 ­0.071[0.484] [0.443] [0.469]

Use of Funds/Quota (if > 1.25), over 2 years (δ1) 0.190 0.070 0.087[0.127] [0.143] [0.131]

Interaction with Centralization Variable (δ3) ­0.761** ­0.042 ­0.168[0.298] [0.222] [0.243]

Extended Fund Facility/Quota, over 2 years (δ1) 0.312*** 0.305*** 0.220**[0.105] [0.117] [0.099]

Interaction with Centralization Variable (δ3) ­1.020* ­0.339* ­0.705[0.525] [0.196] [0.522]

# of reschedulings/restructurings, over 2 years (δ1) 0.584** 0.558** 0.574**[0.267] [0.283] [0.267]

Interaction with Centralization Variable (δ3) ­0.190 ­0.114 ­0.093[0.454] [0.430] [0.451]

Economic and Demographic Controls Yes Yes YesContinent Dummies Yes Yes YesYear Dummies Yes Yes Yes

Constraint on Executive dummy 1 uses Polity IV data: constraint of 1 = 1; constraint of 2­7 = 0; 1 is greatest constraint, 7 is least constraintConstraint on Executive dummy 2 uses Polity IV data: constraint of 1­2 = 1; constraint of 3­7 = 0Political Rights dummy uses Freedom House data: rights of 6­7 = 1; rights of 1­5 = 0; 1 is most free, 7 is least free

Standard errors in brackets; *** p<0.01, ** p<0.05, * p<0.1Control variables not reported are the number of previous protests, urbanization rate, real per capita GDP, population, and a religious fractionalization index

Centralization and/or Coercive Power Variables

Table 6: Logit Regression, IMF Pressure Index and the Presence of Protests

54

DEPENDENT VARIABLE: Protest Presence (0 or 1)

(7.1) (7.2) (7.3)

Constraint on ExecutiveDummy 1

Constraint on ExecutiveDummy 2 Political Rights Dummy

IMF Pressure Indices

Extended Fund Facility/Quota (δ1A) 0.279** 0.312** 0.197*[0.116] [0.127] [0.110]

Interaction with Centralization Variable (δ3A) ­0.865 ­0.434* ­0.576[0.530] [0.236] [0.532]

# of reschedulings/restructurings (δ1B) 0.492* 0.509* 0.553**[0.265] [0.286] [0.268]

Interaction with Centralization Variable (δ3B) 0.126 0.068 ­0.023[0.487] [0.448] [0.472]

Extended Fund Facility/Quota, over 2 years (δ1A) 0.292*** 0.279** 0.196*[0.107] [0.119] [0.101]

Interaction with Centralization Variable (δ3A) ­1.040* ­0.316 ­0.72[0.548] [0.197] [0.543]

# of reschedulings/restructurings, over 2 years (δ1B) 0.496* 0.457 0.504*[0.274] [0.292] [0.274]

Interaction with Centralization Variable (δ3B) ­0.017 0.003 0.025[0.467] [0.436] [0.460]

Economic and Demographic Controls Yes Yes YesContinent Dummies Yes Yes YesYear Dummies Yes Yes Yes

Constraint on Executive dummy 1 uses Polity IV data: constraint of 1 = 1; constraint of 2­7 = 0; 1 is greatest constraint, 7 is least constraintConstraint on Executive dummy 2 uses Polity IV data: constraint of 1­2 = 1; constraint of 3­7 = 0Political Rights dummy uses Freedom House data: rights of 6­7 = 1; rights of 1­5 = 0; 1 is most free, 7 is least free

Standard errors in brackets; *** p<0.01, ** p<0.05, * p<0.1

Table 7: Logit Regression, IMF Pressure Index and the Presence of Protests

Centralization and/or Coercive Power Variables

Control variables not reported are the number of previous protests, urbanization rate, real per capita GDP, population, and a religious fractionalization index

55

(8.1) (8.2)

Centralization and/or Coercive Power Variables Centralization and/or Coercive Power Variables

Constraint on Executive Dummy 1 (η2) 0.234 Constraint on Executive Dummy 1 (η2) ­0.173[1.025] [1.185]

Constraint*EFF/Quota (η3A) ­1.004 Constraint*EFF/Quota, 2 years (η3A) ­2.273[0.769] [2.033]

Constraint*# of reschedulings (η3B) 2.794* Constraint*# of reschedulings, 2 years (η3B) 3.521**[1.637] [1.462]

Constraint on Executive Dummy 2 (η2) 1.676* Constraint on Executive Dummy 2 (η2) 0.109[0.948] [1.226]

Constraint*EFF/Quota (η3A) ­0.264 Constraint*EFF/Quota, 2 years (η3A) ­0.411[0.421] [0.396]

Constraint*# of reschedulings (η3B) 0.795 Constraint*# of reschedulings, 2 years (η3B) 3.310**[1.150] [1.287]

Political Rights Dummy (η2) ­0.142 Political Rights Dummy (η2) ­2.482*[0.998] [1.362]

Political Rights*EFF/Quota (η3A) ­0.490 Political Rights*EFF/Quota, 2 years (η3A) ­1.738[0.667] [1.582]

Political Rights*# of reschedulings (η3B) 1.688 Political Rights*# of reschedulings, 2 years (η3B) 4.352***[1.208] [1.419]

Economic and Demographic Controls Yes Economic and Demographic Controls YesContinent Dummies Yes Continent Dummies YesYear Dummies Yes Year Dummies Yes

Table 8: Logit Regression, Centralization and the Severity of Protests

DEPENDENT VARIABLE: Most Severe Protest in a Given Year (1 or 2/3)

Standard errors in brackets; *** p<0.01, ** p<0.05, * p<0.1Control variables not reported are the number of previous protests, urbanization rate, real per capita GDP, population, and a religious fractionalization index

Constraint on Executive dummy 1 uses Polity IV data: constraint of 1 = 1; constraint of 2­7 = 0; 1 is greatest constraint, 7 is least constraintConstraint on Executive dummy 2 uses Polity IV data: constraint of 1­2 = 1; constraint of 3­7 = 0Political Rights dummy uses Freedom House data: rights of 6­7 = 1; rights of 1­5 = 0; 1 is most free, 7 is least free

56

(9A.1) (9A.2) (9A.3)

# of reschedulings, over2 years = 0

# of reschedulings, over2 years = 1

# of reschedulings, over2 years = 2

EFF/Quota, over 2 years = 0 ­0.036 0.678*** 0.935***[0.249] [0.208] [0.067]

EFF/Quota, over 2 years = 1.25 ­0.406 0.092 0.739*[0.272] [0.461] [0.399]

EFF/Quota, over 2 years = 1.5 ­0.446 ­0.010 0.650[0.273] [0.439] [0.571]

EFF/Quota, over 2 years = 2 ­0.514* ­0.160 0.420[0.276] [0.328] [0.902]

EFF/Quota, over 2 years = 2.5 ­0.573** ­0.256 0.183[0.276] [0.245] [0.928]

EFF/Quota, over 2 years = 3 ­0.627** ­0.326 0.006[0.273] [0.229] [0.662]

EFF/Quota, over 2 years = 4 ­0.722*** ­0.441* ­0.171[0.257] [0.268] [0.261]

EFF/Quota, over 2 years = 5 ­0.800*** ­0.551* ­0.269[0.229] [0.306] [0.273]

Standard errors in bracketsAll control variables reported at the mean, except for the continent dummy (C. America) and year (1985)*** p<0.01, ** p<0.05, * p<0.1

Marginal Effect: ΔPr(Protest Severity = 2 or 3) /ΔCentralization / Coercive Power Variable

Table 9A: Marginal Effects of Centralization Parameter, Regression 8.2

Centralization / Coercive Power Variable = Constraint on Executive Dummy 1

57

(9B.1) (9B.2) (9B.3)

# of reschedulings, over2 years = 0

# of reschedulings, over2 years = 1

# of reschedulings, over2 years = 2

EFF/Quota, over 2 years = 0 0.019 0.623*** 0.923***[0.214] [0.208] [0.098]

EFF/Quota, over 2 years = 1.25 ­0.082 0.585*** 0.911***[0.249] [0.210] [0.097]

EFF/Quota, over 2 years = 1.5 ­0.105 0.575*** 0.908***[0.259] [0.215] [0.097]

EFF/Quota, over 2 years = 2 ­0.153 0.552** 0.900***[0.279] [0.227] [0.099]

EFF/Quota, over 2 years = 2.5 ­0.204 0.524** 0.891***[0.300] [0.244] [0.102]

EFF/Quota, over 2 years = 3 ­0.256 0.493* 0.879***[0.322] [0.264] [0.107]

EFF/Quota, over 2 years = 4 ­0.358 0.417 0.848***[0.366] [0.315] [0.126]

EFF/Quota, over 2 years = 5 ­0.358 0.325 0.804***[0.412] [0.380] [0.164]

Standard errors in bracketsAll control variables reported at the mean, except for the continent dummy (C. America) and year (1985)*** p<0.01, ** p<0.05, * p<0.1

Table 9B: Marginal Effects of Centralization Parameter, Regression 8.2

Marginal Effect: ΔPr(Protest Severity = 2 or 3) /ΔCentralization / Coercive Power Variable

Centralization / Coercive Power Variable = Constraint on Executive Dummy 2

58

(9C.1) (9C.2) (9C.3)

# of reschedulings, over2 years = 0

# of reschedulings, over2 years = 1

# of reschedulings, over2 years = 2

EFF/Quota, over 2 years = 0 ­0.353 0.314 0.878***[0.239] [0.231] [0.169]

EFF/Quota, over 2 years = 1.25 ­0.481* ­0.031 0.550[0.259 [0.180] [0.534]

EFF/Quota, over 2 years = 1.5 ­0.500* ­0.067 0.457[0.262] [0.166] [0.612]

EFF/Quota, over 2 years = 2 ­0.537** ­0.119 0.280[0.267] [0.145] [0.653]

EFF/Quota, over 2 years = 2.5 ­0.571** ­0.156 0.144[0.273] [0.145] [0.541]

EFF/Quota, over 2 years = 3 ­0.604** ­0.184 0.057[0.279] [0.160] [0.373]

EFF/Quota, over 2 years = 4 ­0.666** ­0.236 ­0.023[0.287] [0.208] [0.140]

EFF/Quota, over 2 years = 5 ­0.723** ­0.289 ­0.054[0.290] [0.267] [0.090]

Standard errors in bracketsAll control variables reported at the mean, except for the continent dummy (C. America) and year (1985)*** p<0.01, ** p<0.05, * p<0.1

Table 9C: Marginal Effects of Centralization Parameter, Regression 8.2

Marginal Effect: ΔPr(Protest Severity = 2 or 3) /ΔCentralization / Coercive Power Variable

Centralization / Coercive Power Variable = Political Rights Dummy

59

C Figures

Figure 1: Three Cases, �� i�2 plane

Figure 2: Case 3, Graphically

60

Figure 3: Three Cases, Qualitatively

Figure 4: Total Protests by Year

61

D Protest Data Description

I �rst identi�ed all countries that either satis�ed at least one of the three "IMF pressure" variables noted in

the text or were identi�ed by Walton and Seddon (1994) as having protests in this period. For all of these

countries, I searched all articles in Lexis-Nexis Academic that arose from the following search string: (IMF

OR austerity) AND (riot OR protest) AND [country]. This method produced 116 instances of protest, as

opposed to 146 instances found by Walton and Seddon. The difference is likely a result of coding of small

protests � I did not count events such as general strikes (unless violence broke out) or small, sector-speci�c

strikes; it is not clear where Walton and Seddon's cutoff point occurs. However, there do not appear to be any

inconsistencies in the counting of large protests.

As noted in Section 3.1, protests and riots were only documented if they resulted from austerity measures

or IMF pressure. In the case of protests, a speci�c aim of those protesting had to be some sort of austerity

measure, such as higher prices or reduced wages. If riots broke out, they are only counted in the data set if

they were a result of some austerity measure. Thus, a government coup would not be part of the universe of

observations, unless it were directly the result of austerity measures.

As noted, I subjectively coded each of these protests by their severity, using the following criterion. An

instance was scored (1) if it was a small, con�ned (to one or two cities) protest or general strike, (2) if it

was either prolonged but con�ned or widespread but not prolonged, with minimal deaths resulting, and (3)

if the protest was prolonged, contained widespread riots, and resulted in many deaths. The severity of some

protests was not obvious. For these protests, I created an "alternate index", listed below. The index does not

take a value if a protest cannot reasonably be considered of a different severity than is given in the severity

index. The alternate takes on a different severity value if it could reasonably be considered a different severity

than the one I originally coded. All protests are listed below with a brief description and severity level.

62

Country Month Year Description SeverityIndex

AlternateIndex?

February 1989 Student protests after gas price increase; 1000 involved in fightswith police 1 ­­

September 1991 30 dead; precipitated regime change 3 2

Côte d'Ivoire February 1990 Protests against wage cuts; lasted two months until austeritydropped; relatively non­violent

3 2

Gabon December 1989 Series of strikes against austerity; national conference called 1 ­­

Ghana November 1978 80 strikes after austerity measures 1 ­­

Mozambique January 1990 Waves of strikes and protests over austerity 1  ­­

Niger February 1990 10 students killed, 50 wounded in protest against austerity 2 ­­

April 1988 6 students killed in riot over increased gas prices 2 1May 1989 Two weeksof rioting against austerity, 30 killed 2 ­­May 1992 23­100 dead in riots in many cities 2 ­­

Senegal March 1988 Urban riots over austerity measures 1 ­­

Sierra Leone July 1980 4 days of riots after gas price hike; 2 dead 2 1

March 1981 Protests in northern towns over inflation, shortages 1 ­­

January 1982 4 days of student protests over austerity; thousands involved, 28deaths

2 ­­

March 1985 3 days of rioting following food price hikes; 5 deaths, 2000arrested 2 ­­

October 1987 Student protests over food and fuel price hikes, 10,000 involved,one death

2 1

December 1987 Widespread demonstrations after price hikes 1 ­­

December 1986 15 dead, 1000 arrested in widespread riot over hike in price ofcornmeal; price hike rescinded 2 ­­

January 1989 61 arrested after change in food subsidies 1 ­­July 1989 Youth food riots after food price hike 1 ­­June 1990 30 dead, hundreds injured in weekend riot 3 2

Country Month Year Description SeverityIndex

AlternateIndex?

Bangladesh June 1986 Thousands protest in Dhaka over rising cost of living 1 ­­

February 1986 Thousands protest price hikes 1 ­­September 1991 Thousands march in New Delhi to protest austerity 1 ­­November 1991 Millions strike against economic policies 1 ­­June 1992 Millions strike against economic policies 1 ­­September 1992 Protests in New Delhi after oil price hike 1 ­­November 1992 1,200 arrested, 50 injured in protest over rising prices 1 ­­

November 1984 8,000 protest against IMF in Manila 1 ­­July 1991 Violent transport driver strike of economic policies 1 ­­January 1992 Protest against rise in electricity prices 1 ­­

South Korea May 1980 Series of strikes and protests over austerity; 300 dead 3 2

Asia

India

Philippines

Nigeria

Sudan

Zambia

Africa

Congo (Dem.Rep.)

63

Country Month Year Description SeverityIndex

AlternateIndex?

April 1984 Protests over entire month throughout D.R. against austerity; 120dead, 6,000 arrested 3 ­­

July 1984 25,000 protest resumption of IMF talks 1 ­­August 1984 Students protest price hikes; 1 dead 1 ­­November 1984 Hundreds protest milk price hikes 1 ­­January 1985 Waves of protests after price hikes; hundreds arrested 1 ­­March 1988 Two weeks of protests over economic policies; 4 dead 2 1August 1990 2 day strike over price increases; 8 dead; 1000 arrested 2 ­­

January 1986 8,000 protest over austerity measures 1 ­­February 1986 50,000 protest over austerity measures 1 ­­

Guatemala September 1985 Protests over bus fare increase; 1 dead, 466 arrested 2 1

Haiti June 1984 Violence in 5 towns over 3 months over austerity; 3 dead 2 1

Honduras May 1984 20,000 protest against austerity in Tegucigalpa 1 ­­

January 1979 3 days of rioting over gas price increase; 7 dead 2 ­­January 1985 Nationwide riots over gas price increase; 10 dead 2 ­­April 1985 Protests in several towns over austerity 1 ­­March 1986 Student, opposition protests over austerity 1 ­­

February 1983 Thousands protest in Mexico City against austerity 1 ­­June 1983 Thousands of students protest against austerity 1 ­­October 1983 Thousands protest in Acapulco against austerity 1 ­­October 1985 7,000­10,000 marched against austerity 1 ­­

Nicaragua October 1990 Protests throughout Managua over austerity 1 ­­

November 1984 150,000 marched in protest over austerity 2 1July 1985 Violent 2­day general strike over austerity 1 ­­March 1986 Week­long strike, riot against austerity; one death 2 1June 1991 Students, teachers riot over austerity in Panama City 1 ­­

Country Month Year Description SeverityIndex

AlternateIndex?

Bulgaria November 1990 20,000 protest austerity plan 1 ­­

January 1990 Street protests in Warsaw over austerity 1 ­­April 1992 30,000 protest plans for austerity measures 1 ­­May 1992 80 protests throughout country over IMF plans 1 ­­

November 1987 Thousands of workers riot over wage cuts 1 ­­December 1987 Student and worker protests throughout country 1 ­­December 1989 Hundreds dead in widespread riots 3 ­­April 1991 15,000 demonstrate in Bucharest 1 ­­September 1991 Miners, protesters battle police for 3 days; 3 dead 3 2

Romania

Panama

Europe

Poland

DominicanRepublic

El Salvador

Jamaica

Mexico

Central America/Caribbean

64

Country Month Year Description SeverityIndex

AlternateIndex?

November 1986 3 days of student rioting in Constantine; 4 deaths; 186 jailed 2 1

October 1988 Followed increase in food prices; over a week of riots; 159 dead,154 wounded; riots across country

3 ­­

Egypt January 1977 Riots throughout country over food price increase; 79 dead 3 ­­

Iran August 1991 Sporadic protests over bus fare increases 1 ­­

Jordan April 1989 Riots for a week throughout country over austerity; 8 dead 3 2

July 1981 General strike after price increases; 66 killed 3 2

January 1984 Riots in half dozen cities after food price hikes; 240 killed overfour days 3 2

December 1990 33 killed, 210 arrested in two days of riots over economichardships worsened by austerity

2 ­­

Tunisia January 1984 Riots in main towns after bread price doubled 1 ­­

Country Month Year Description SeverityIndex

AlternateIndex?

Argentina March 1982 6 wounded, 400 arrested in austerity protest 2 1

March 1983 Strike by public workers against austerity; 2 dead 1 ­­September 1985 Month­long strike; state of siege declared; thousands arrested 2 ­­January 1986 Waves of protests against austerity measures 1 ­­August 1986 Widespread protests; state of siege declared 2 1April 1987 10,000 demonstrate against austerity 1 ­­November 1987 10,000 demonstrate against austerity 1 ­­November 1989 Teachers strike, 700 arrested, state of siege declared 1 ­­

May 1983 Riots in Sao Paulo and Rio de Janeiro over austerity 2 1July 1983 Workers around the country strike against austerity 2 1September 1983 Poor riots in Rio de Janeiro; austerity behind price rise 1 ­­October 1985 Thousands protest non­repayment of government debt 1 ­­

November 1986 Thousands riot in Brasilia against austerity; looting andvandalizing 2 1

July 1987 Violent demonstrations in Rio de Janeiro after increase in busfares; 100 buses set on fire; military called in

2 1

March 1989 Nationwide protest over wage freeze 1 ­­

March 1990 Miners rampage in Peixoto; 2 dead, many injured; protest overausterity measures

2 ­­

June 1983 Labor protests and riots; 4 dead, hundreds arrested 2 1August 1983 Riot police clash with protesters; 200 arrested 1 ­­

October 1983 3 days of protests; 40,000 in Santiago; more protests elsewhereover austerity; 6 dead 2 ­­

July 1985 Riot police break up anti­austerity protest in Santiago; 70 arrested,3 bombs dismantled

2 1

October 1982 3 days of demonstrations against austerity; state of emergencycalled

2 1

January 1985 Street protests, bus driver strike over raise in fuel prices 1 ­­January 1986 Protests over prices in front of U.S. Embassy 1 ­­September 1992 Protests in 3 cities over austerity; 3 explosions 1 ­­

July 1976 Demonstrations against tax and price increases; civil rightssuspended

2 1

June 1977 Students demonstrate against price rise in 5 cities 2 1July 1977 Students, workers protest price hikes; 2 dead; 12 cities 2 ­­May 1978 Food price riots due to austerity; 12 dead 2 ­­May 1978 Youths revolt throughout Peru; 3 dead 2 ­­March 1983 24­hour strike against austerity; minimal violence 1 ­­September 1983 24­hour strike against austerity; 2 dead; 100 arrested 2 1March 1984 Workers and students strike in 6 towns; 1 dead 2 1November 1984 Student protest over austertiy in Lima; 250 arrested 1 ­­September 1988 Riots against austerity; looting and windows smashed 1 ­­August 1990 Week­long riots over food price hikes; 3 dead 2 ­­July 1991 All­out strike national strike; 2 dead 2 1

Chile

Ecuador

Peru

Middle East/North Africa

South America

Bolivia

Brazil

Algeria

Morocco

65

E Robustness Checks

DEPENDENT VARIABLE: Protest Presence (0 or 1)

(A1.1) (A1.2) (A1.3)

Democracy 0 Democracy 1 Democracy 2IMF Pressure Indices

Use of Funds/Quota (if > 1.25) (δ1) ­0.068 ­0.051 ­0.087[0.162] [0.157] [0.148]

Interaction with Centralization Variable (δ3) 0.212 0.199 0.335[0.221] [0.224] [0.228]

Extended Fund Facility/Quota (δ1) 0.127 0.202 0.14[0.165] [0.139] [0.135]

Interaction with Centralization Variable (δ3) 0.100 ­0.031 0.078[0.203] [0.185] [0.188]

# of reschedulings/restructurings (δ1) 0.753** 0.828*** 0.763***[0.320] [0.289] [0.285]

Interaction with Centralization Variable (δ3) ­0.348 ­0.712 ­0.534[0.438] [0.456] [0.458]

Use of Funds/Quota (if > 1.25), over 2 years (δ1) ­0.043 ­0.028 ­0.065[0.154] [0.150] [0.142]

Interaction with Centralization Variable (δ3) 0.145 0.135 0.259[0.214] [0.219] [0.222]

Extended Fund Facility/Quota, over 2 years (δ1) 0.126 0.161 0.12[0.141] [0.125] [0.122]

Interaction with Centralization Variable (δ3) 0.108 0.045 0.129[0.180] [0.168] [0.170]

# of reschedulings/restructurings, over 2 years (δ1) 0.521* 0.512* 0.512*[0.301] [0.268] [0.267]

Interaction with Centralization Variable (δ3) 0.003 ­0.034 0.035[0.428] [0.458] [0.466]

Economic and Demographic Controls Yes Yes YesContinent Dummies Yes Yes YesYear Dummies Yes Yes Yes

Standard errors in brackets; *** p<0.01, ** p<0.05, * p<0.1

Table A.1: Democracy and the Presence of Protests

Centralization and/or Coercive Power Variables

Control variables not reported are the number of previous protests, urbanization rate, real per capita GDP, population, and a religious fractionalization index

All democracy dummies use Polity IV data: democracy 0 = 1 if Polity variable = 0; democracy 1 = 1 if Polity variable = 0 or 1; democracy 2 = 1 if Polity variable = 0, 1,or 2

66

(A2.1)

Democracy 0 (η2) ­2.204**[1.045]

Democracy 0*# of reschedulings (η3B) ­0.258[1.020]

Democracy 0 (η2) ­0.144[1.281]

Democracy 0*# of reschedulings, 2 years (η3B) ­2.854**[1.251]

Democracy 1 (η2) ­2.255**[0.994]

Democracy 1*# of reschedulings (η3B) 1.484[0.926]

Democracy 1 (η2) ­0.51[1.175]

Democracy 1*# of reschedulings, 2 years (η3B) ­0.791[0.945]

Democracy 2 (η2) ­2.248**[1.037]

Democracy 2*# of reschedulings (η3B) 1.23[0.913]

Democracy 2 (η2) ­0.674[1.211]

Democracy 2*# of reschedulings, 2 years (η3B) ­0.756[0.949]

Economic and Demographic Controls YesContinent Dummies YesYear Dummies Yes

Standard errors in brackets; *** p<0.01, ** p<0.05, * p<0.1

Table A.2: Democracy and the Severity of Protests

All democracy dummies use Polity IV data: democracy 0 = 1 if Polity variable = 0; democracy 1 = 1 ifPolity variable = 0 or 1; democracy 2 = 1 if Polity variable = 0, 1, or 2Control variables not reported are the number of previous protests, urbanization rate, real per capita GDP,population, and a religious fractionalization index

Centralization and/or Coercive Power Variables

DEPENDENT VARIABLE: Most Severe Protest in a Given Year (1 or 2/3)

67